Methodology changelog
A citation outlives the rule it was computed under. Every file we publish carries methodology_version — today 1.33.64 — top level in price-index.json and prices.json, and as a column in every CSV and in each weekly receipt, because a CSV row is what survives copying. This page says what each version means. The rules themselves are on the methodology page.
What the parts mean
- Patch — collection and coverage: a provider added or dropped, a parser fixed, a quote withheld. The definition of the published figure is untouched, so figures across a patch are comparable.
- Minor — new fields, a new cut of the same quotes, or a new rule about which reads count as current. Figures published before it stay valid.
- Major — the definition of the published figure itself. It should never appear here: a number that changes meaning gets a new series under its own name instead, so a figure someone cited never quietly becomes a different measurement. If this column ever says major, the series was renamed in the same change.
Changes
| Date | Version | Kind | What changed | Definition of the published figure |
|---|---|---|---|---|
| 1.33.64 | patch | Meshive is read from the gpu_offers array of its public pricing API. Its team clarified that Meshive operates its own fleet in South Korea and that the broader HTML catalog includes cards it does not currently offer. The first API read lists RTX 5090 32 GB and RTX PRO 5000 Blackwell 48 GB, each with on-demand and spot rates per GPU-hour, including default vCPU and RAM. The catalog quotes are retired from current listings. Offer buttons open Meshive’s pricing page, while the listing links the API used to verify the figures. The definition of the index did not change. | unchanged | |
| 1.33.63 | patch | A new card, the RTX 5880 Ada Generation (48 GB), with two sources in its first hourly read at 04:40 UTC: Meshive's catalog table, $0.596 per GPU per hour on demand and $0.357 spot (each a "from" rate), and the Vast.ai marketplace floor, $0.8037 for one GPU (the median of exactly three verified asks, so the quote may read "thin market" on some hours) and $1.6037 for a 2-GPU machine. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.62 | patch | EmpirioLabs renamed the rows of its GPU table on the afternoon of 2026-10-29 by form factor and memory, and three of its quotes stopped being read after 15:40 UTC. They are read again under the new names, at the same figures: "H200 SXM" as the H200 ($4.99), "A100 80GB SXM" as the A100 80GB ($1.70; the new "A100 80GB PCIe" row at $2.10 is not mixed into it) and "V100 16GB", a 4-GPU node, as the V100 ($1.80 a node, $0.45 per GPU per hour). First hourly read 22:40 UTC. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.61 | patch | Meshive (Steppod, LLC, US) is read from the catalog table on its pricing page (meshive.ai/pricing): an on-demand and a spot rate per GPU-hour for each card, which Meshive sets itself on machines that hosts contribute - customers are matched to a host and never see hosts' own asks, so the figure is the rate as printed, not a marketplace floor. The page calls every figure a per-GPU starting price (vCPU and memory above an included amount and storage bill apart), so each carries the "from" tag; US dollars, no location named. Where two forms of an A100 land on one of our cards the cheaper is kept. First hourly read 19:40 UTC: 82 quotes over 41 cards, among them RTX 4060 $0.066 (spot $0.040) and RTX 4070 SUPER $0.185, each card's second source, RTX 3060 Ti $0.154 and RTX 3070 Ti $0.144, each card's third, H100 SXM $3.059, H200 $4.011 and B200 $8.282 per GPU per hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.60 | patch | OpenRelay (formerly VectorLay) is read from its GPU catalog (openrelay.inc/gpus): one hourly list rate per GPU, which OpenRelay sets and bills by the second on a distributed network of contributed hardware - not a marketplace of hosts' own asks, so the figure is the rate as printed, per GPU, US dollars, no location named. First hourly read 16:40 UTC: RTX 4090 $0.35, RTX 3090 $0.18, RTX 5090 $0.50, RTX PRO 6000 $1.55, A100 80GB $1.10, H100 SXM $2.60, B200 $6.30, B300 $8.49, MI355X $6.30 per GPU per hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.59 | patch | TheAI Cloud's pricing page was rebuilt on 2026-10-29 into one rate table of four lots (B300 for 3 months in Kuala Lumpur, B300 for 36 and 60 months in Iowa, B200 month-to-month in Los Angeles from November 2026), and the 15:37 UTC read failed on it. The module reads the new table for the same quote as before, the 36-month B300 node; the other lots have no market on this site or are not sold yet. First hourly read 16:40 UTC: $37.60 an 8-GPU node, $4.70 per GPU per hour, unchanged. The definition of the published figure did not change. | unchanged | |
| 1.33.58 | patch | TensorCloud (Tensordata Global Private Limited, India) is read from the GPU cards on its home page (tensordata.com, the section its Pricing link opens): one GPU per card, in rupees as printed - the page says nothing about GST either way. A card the page marks Sold Out is not a quote (the H100 on 2026-10-29). First hourly read 10:40 UTC: A100 SXM4 80 GB 126 rupees an hour, $1.3232 at the ECB rate of the day. No other figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.57 | patch | Lium's nodes feed is read for the 32 GB V100: Lium files both V100s under one name, "Tesla V100 Tensor Core GPU", and the module had a window for the 16 GB card only. Nodes reporting 28-36 GB a GPU are now the V100 32GB, read by the marketplace rule every Lium card follows (the median of the five cheapest verified asks; a whole 8-GPU node in the Multi-GPU machines book). First hourly read 07:40 UTC: $3.20 for an 8-GPU node, $0.40 per GPU per hour. No other figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.56 | patch | Krutrim Cloud is read from its pricing page (www.olakrutrim.com/pricing) instead of its compute documentation, which on 2026-10-28 dropped the price columns from its GPU table; the GPU Instances table there prints the same on-demand rates in rupees for a whole VM - A100 80 GB 189 rupees and H100 213 rupees an hour - and the price link of both quotes now goes to that page. The Monthly, 6-month and 1-year commitment columns and the AI Pods table are not read. No figure changed, and no other provider's. The definition of the published figure did not change. | unchanged | |
| 1.33.55 | patch | EmpirioLabs' GPU table is read for three more rows: the T4 ("16 GB VRAM") and the RTX PRO 5000 Blackwell ("48 GB VRAM", the 48 GB edition), one card a row, and the V100, which the table prices as a "64 GB VRAM · 4-GPU node" - four 16 GB cards, stored per node with its count, as the module already reads every "N-GPU node" row. First hourly read 04:40 UTC: T4 $0.35, RTX PRO 5000 Blackwell $2.00 per GPU per hour, V100 node $1.80 ($0.45 per GPU). No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.54 | patch | MWS Cloud Platform (MTS Web Services, Russia) is read: the GPU line of its public tariff page. A GPU virtual machine is billed there as separate lines - the card, each vCPU and each gigabyte of memory - so the figure is the card's line alone, the processor and memory added on top, as Yandex Cloud's is read. Hourly column, in roubles with 22% VAT as its header states, converted at the Bank of Russia's rate of the day; a line whose footnote or region dates it after today is not read yet. First hourly read 01:40 UTC: NVIDIA A100 80GB 263.5507 RUB an hour in ru-central1, $3.2591 per GPU per hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.53 | patch | EmpirioLabs' GPU table is read for three more rows it began to print: the RTX 5080, RTX 5070 Ti and RTX 5060 Ti, each "16 GB VRAM", one card a row, read per GPU on demand like the rest of the table. The 16 GB the page states is what puts the RTX 5060 Ti on the 16 GB card; the 8 GB edition has no id here. First hourly read 22:40 UTC: RTX 5080 $0.80, RTX 5070 Ti $0.60, RTX 5060 Ti $0.40 per GPU per hour. Its "V100 | 64 GB VRAM · 4-GPU node" row is not read yet. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.52 | patch | Yotta Labs is read: the GPU table on its pricing page, one list price per card with the pod's memory, RAM and vCPUs, in two columns, On-demand and Spot. Each row is read per GPU as an instance; a spot cell at or above the same row's on-demand rate is not read (the A100 PCIe 80G, 1.1664 against 0.92), and a bare "H100" is listed as the H100 with its form factor not stated. First hourly read 16:39 UTC: on demand RTX 4090 $0.48, RTX A6000 $0.45, RTX 5090 $0.70, A100 80GB $0.92, RTX 6000 Ada $0.97, RTX PRO 6000 $1.35, H100 $2.56, H200 $2.50, B200 $5.37, B300 $7.64; spot RTX A6000 $0.27, A100 80GB $0.5724, RTX 6000 Ada $0.3888, RTX PRO 6000 $0.7236, H100 $0.95, H200 $1.60, B200 $2.20, B300 $3.10 per GPU per hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.51 | patch | Octa.space's "RTX 4000 (20 GB)" is read as the RTX 4000 Ada. Both the RTX 4000 Ada and the RTX 4000 SFF Ada carry 20 GB, so the name does not tell them apart; each machine card prints its power limit and its maximum ("Power Limit 105W (130W)"), and the RTX 4000 Ada is a 130 W card while the SFF Ada is a 70 W one. A machine is counted only when every maximum it prints is 100 W or more; any other stays unread. First hourly read 23:40 UTC on 2026-10-27: a floor of $0.22 per GPU per hour (node 1073), on the same verified-machine rule and trimmed floor as every Octa.space figure, inside the card's live on-demand range of $0.20 to $0.87. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.50 | patch | Helios Cloud now prints each reserved rate as a range ("USD per GPU-hour, shown as a published range", for example $8.17 to $8.74 for a GB300 on one year) and does not say what places a buyer within it. Each rate is read at the lower end of its range and listed as a "from" figure; the upper end is not listed. The hourly read of 2026-10-27 21:37 UTC failed on the new page, so Helios' rates went unconfirmed for an hour. First hourly read 22:39 UTC: GB300 from $8.17 (1 year) and $5.10 (3 years), B300 from $7.00 and $4.37, RTX PRO 6000 from $1.35 and $1.23 per GPU per hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.49 | patch | One more marketplace is read: Octa.space, where hosts set their own price per machine. Its figure is a floor, as on Vast.ai, Clore.ai and Lium: the median of the five cheapest asks on machines whose Stability and 30-Day Machine Uptime, as Octa.space prints them, are both 95% or more and which have at least one successful rental (Octa.space prints no verified badge; the thresholds are ours), withheld when fewer than three qualify. The price is the machine's Total Cost as its card shows it, the default 100 GB disk included. Machines rented whole are a separate row; machines mixing two GPU models are not read. First hourly read 16:40 UTC: RTX 3090 $0.26, RTX 4090 $0.379, RTX 5070 $0.29, RTX 5080 $1.10, RTX 5090 $0.521 per GPU per hour, and a whole 2x RTX 3090 machine at $1.25 an hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.48 | patch | One more provider is read: STN, which sells reserved GPU capacity only, with a minimum contract of three years; its three-year rate per GPU-hour is listed as a 3-year committed rate. The 4- and 5-year rates are not listed, nor its Vera Rubin rate, which the page calls indicative; the GB300 rate is a "from" figure and is listed as one. First hourly read 10:39 UTC: B200 $5.10, B300 $5.85, GB300 NVL72 from $7.15 per GPU per hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.47 | patch | One more provider is read: Helios Cloud, which sells reserved GPU capacity only, priced per GPU-hour for terms of one to five years; its 1-year and 3-year rates are listed as committed rates, the GPU alone. The 2-, 4- and 5-year rates are not listed. First hourly read 10:39 UTC: GB300 $6.77 (1 year) and $6.04 (3 years), B300 $5.89 and $5.25, RTX PRO 6000 $1.44 and $1.29 per GPU per hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.46 | patch | Nebula Block's RTX A4000 is read: the row had been out of stock since the provider was added and is back in stock. First hourly read 22:40 UTC: $0.17 per GPU per hour, on-demand. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.45 | patch | One more provider is read: Airon, which rents bare-metal GPU servers by the hour and prints each server's price as a "from" figure; listed as starting prices, per GPU, from the smallest server of each card. The headline prices at the top of its page are not listed, because they do not name the server they apply to, and neither are its eight-GPU H100 and B200 servers, because the page describes those cards two different ways. First hourly read 19:40 UTC: H100 PCIe from $2.00, RTX A6000 from $0.60, RTX A5000 from $0.45 per GPU per hour; two L40S from $1.80 an hour ($0.90 per GPU); eight H200 from $21.04 an hour ($2.63 per GPU). No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.44 | patch | One more provider is read: Cato Digital, which rents whole bare-metal servers by the hour, eight GPUs each, priced per server and shown here per GPU. One server is listed per card, the smallest; the monthly prices beside the hourly ones assume a year paid in advance and are not listed. Its servers are "fail in place" - a failed server is replaced, its storage is not - which is not preemption, so the rates are listed as on-demand. First hourly read 13:40 UTC: eight V100 32GB $11.20 an hour ($1.40 per GPU), eight V100 16GB $6.63 ($0.83 per GPU), eight A100 80GB $12.00 ($1.50 per GPU). No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.43 | patch | One more provider is read: Runware, an inference platform whose pricing page also rents its GPUs to run your own models and containers, billed per second of runtime; listed as serverless rates. Only the hourly column is read, and only where it matches the per-second price times 3,600; the reserved "as low as" rates name no term and are not listed. Its H100 is listed without a form factor, which the page does not print. First hourly read 13:40 UTC: L40S $1.60, RTX PRO 6000 $1.99, H100 $3.95, H200 $4.50, B200 $6.00, B300 $7.00 per GPU per hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.42 | patch | One more provider is read: Lightning AI, which rents cloud development machines billed by the second in its own credits, one dollar each. Only its machines with a single GPU are listed - its free tier rents them without a subscription, while machines with more GPUs need a paid monthly plan - and only at the regular rate; its interruptible rate is printed as a range and is not listed. Its RTX PRO 6000 is not listed, because the page does not say which version of the card it is, and its H100 is listed without a form factor. First hourly read 02:40 UTC: T4 $0.55, L4 $0.79, L40S $3.54, A100 40GB $2.19, A100 80GB $1.85, H100 $5.45, H200 $4.50 per GPU per hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.41 | patch | One more provider is read: Nebula Block, a GPU cloud with its own data centres in Canada. Its pricing page draws its instance table with a script, so the page is read as a browser shows it. Instances whose name ends in "spot" are listed as spot rates and the others as on-demand; instances marked out of stock are not listed, and its H100 and H200 instances are not, because the page does not say which form of the card they are. First hourly read 01:40 UTC: A100 80GB $1.43, L40 $1.06, RTX 5090 $1.02, RTX A6000 $0.54 per GPU per hour on demand; L40S $1.06 and an eight-GPU B200 machine $44.55 an hour ($5.57 per GPU) as spot. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.40 | patch | One more card is tracked: the Intel Gaudi 2, an AI accelerator with 96 GB of memory and the first Intel part on this site. Two providers we already read print an hourly rate for it, and both were read in the same hour: Denvr Dataworks, per accelerator, and Sesterce, for a machine of eight. First hourly read 22:40 UTC: Denvr $1.10 per accelerator per hour; Sesterce $9.68 an hour for eight, $1.21 each, listed as a starting price like its other rows. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.39 | patch | One more provider is read: airgpu, which rents cloud PCs streamed to your device, paid from bought credit with no subscription. Its data-centre cards are priced "from" a figure across data centres worldwide and are listed as starting prices; its RTX cards are one fixed machine each, available in North America, and are listed with that region. Rates are in US dollars excluding VAT, as the page prints them. First hourly read 17:39 UTC: T4 from $0.65, L4 from $0.90, A10G from $1.05, L40S from $1.20; Quadro RTX 4000 $0.65, RTX A4000 $1.10, RTX A5000 $1.55 (North America) per GPU per hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.38 | patch | One more provider is read: Vagon, which rents cloud computers - a Windows workstation streamed to the browser - billed by the minute from a balance. Each card comes in several sizes with more cores and memory; the size with the lowest price per GPU is listed, as the whole computer. Its storage plan is charged beside the rate and is not added to it. First hourly read 16:39 UTC: A10G $1.67 (Spark), T4 $0.99 (Planet) per GPU per hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.37 | patch | One more card is tracked: the NVIDIA GB10 Grace Blackwell superchip, the one GPU in an NVIDIA DGX Spark, with 128 GB of memory shared between CPU and GPU. Two providers print an hourly rate for one DGX Spark, so the machine's rate is the GPU's: Enverge and GPUwerk. GPUwerk's linked two-Spark cluster and its partner-hosted Spark are not listed, since each would be a second figure for the same quote. First hourly read 04:39 UTC: Enverge $0.75, GPUwerk $0.79 per GPU per hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.36 | patch | One more provider is read: GPUwerk, an EU operator that rents dedicated machines by the minute from a prepaid balance. Its pricing page lists each x86 card with its status; only a card marked available is read, so its L40S, listed as planned with a waitlist, is not. Its DGX Spark machines (GB10) are not a card we track and are not listed. Rates are in US dollars excluding applicable tax, as the page prints them. First hourly read 01:39 UTC: RTX 5090 $0.89 per GPU per hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.35 | patch | One more provider is read: CloudGPU, a platform in public beta that rents single GPUs by the minute and prints one hourly list price per card. Its machines are suppliers' servers, but the price is the platform's own, so each figure is read as a list price, not as a market floor. China-market cards we do not track (A800, H800, H20, RTX 4090D, RTX 6000D, Ascend 910B) and its 48 GB RTX 4090 are not listed. First hourly read 19:40 UTC: RTX 5090 $0.59, RTX 4090 $0.33, RTX 3090 $0.21, V100 32GB $0.24, A100 80GB $1.49, A100 40GB $0.63, A30 $0.22, A40 $0.39, RTX A6000 $0.44, RTX PRO 6000 $1.49 per GPU per hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.34 | patch | One more provider is read: Tata Communications (Vayu AI Cloud), an Indian operator whose GPU page prints "Starting from" hourly rates in rupees for 8-GPU bare-metal servers and for GPU virtual machines. A bare-metal server is read per node with its eight GPUs, a virtual machine at its smallest size per GPU, all marked as "from" figures. Its virtual-machine H200 names no form factor on a page that also sells the SXM and the NVL card, so it is not listed. First hourly read 16:39 UTC: 8x H100 SXM INR 3,905 a node ($5.14 per GPU per hour), 8x H200 SXM INR 4,247 ($5.59), 8x H200 NVL INR 4,122 ($5.43), L40S INR 200 ($2.11), L4 INR 90 ($0.95), at the ECB rate of 23 October. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.33 | patch | One more provider is read: FarmGPU, which rents 8-GPU clusters on demand and prints one hourly figure per cluster; the page's own comparison line gives that figure for the H100 "Per GPU/hr", so each card is read per GPU. Long-term terms, quoted per cluster, are not read. First hourly read 16:39 UTC: H100 SXM $3.29, H200 $4.59, B200 $6.79, B300 $7.89 per GPU per hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.32 | patch | One more provider is read: FluidCore, an Indian operator that rents whole multi-GPU bare-metal nodes and prints their rates in rupees in a table on its pricing page. The on-demand node-hour is read, per node with its GPUs; where a card comes in two node sizes the lower price per GPU is kept. The 1- and 3-month reserved columns are not read. First hourly read 11:40 UTC: 8x A100 80GB INR 1,850 a node ($2.44 per GPU per hour), 8x H100 SXM INR 4,200 ($5.53), at the ECB rate of 23 October. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.31 | patch | One more provider is read: FlexAI, whose pricing page prices dedicated GPUs per GPU-hour, on demand, beside per-token model prices that are not read. First hourly read 10:40 UTC: B200 $6.25, H200 $3.15, H100 $2.10 per GPU per hour, the GPU alone as printed. Its A100 names no memory, so which A100 it is cannot be told and it is not listed. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.30 | patch | One more provider is read: Limestone Networks, which rents whole dedicated 8-GPU bare-metal nodes and prints each node's hourly and monthly rate on its pricing page. The node-hour is read with its eight GPUs. First hourly read 10:40 UTC: RTX 5090 $1.25, RTX PRO 6000 $2.95, A100 80GB SXM $2.25, H100 SXM $2.95, B300 $6.25 per GPU per hour. Its GB300 rack prints no hourly rack rate and is not listed. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.29 | patch | Sesterce's B200 is read again: its console's offer table prints "B200 1x $7.17/h 192 GB", and the parser had set the name aside until it was matched to one card. 192 GB is the B200, the memory other sources of the card print. First hourly read 07:40 UTC: $7.17 per GPU per hour, on demand, a "from" figure as every Sesterce row; the quote last stood at $4.11, read from Sesterce's old home page until 8 October. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.28 | patch | Northflank's H200 and B200 are read: its pricing page added both rows to its GPU table on 22 October, per GPU per hour like the cards already read there, and the parser had set them aside until each name was matched to one card. "H200 141GB" is the H200 (not the NVL edition), "B200 180GB" the B200. First hourly read 22:40 UTC: H200 $3.14, B200 $5.87 per GPU per hour; Northflank's five other figures did not change. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.27 | patch | One more provider is read: TheAI Cloud, which sells a whole 8-GPU B300 node on reserved terms only and prints its rates in a table on its pricing page. The 36-month rate is read as a three-year commitment, per node with its eight GPUs; the 24-month rate has no market on this site and is not listed, as at Civo. First hourly read 16:40 UTC: B300, 36 months, $37.60 a node, $4.70 per GPU per hour, take-or-pay with 30% at signing. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.26 | patch | Crusoe is read again after three hourly runs without rows (19:37 to 21:37 UTC): its pricing page now sets the GPU table out as three columns side by side - models, on-demand, spot - and the parser pairs them by position, stopping if they differ in length. First hourly read 22:39 UTC: the same six rows as before, B300 $8.96 per GPU per hour (was $9.90), H200 $4.29, H100 $3.90, A100 80 GB $2.00, L40S $1.50, MI300X $3.45. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.25 | patch | One more Crusoe row is read (first hourly read, 13:40 UTC). Its GPU instances table now prints an on-demand rate for the B300 with 288 GB, $9.90 per GPU per hour; the spot cell still reads "Contact sales". No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.24 | patch | One more Sesterce row is read (first hourly read, 09:40 UTC). Its console lists a GH200 with 96 GB, one GPU on demand at $2.52 an hour in US dollars as printed; the row is not on the console every hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.23 | patch | One more Sesterce row is read (first hourly read, 05:40 UTC). Its console lists an RTX 4000 Ada with 20 GB, one GPU on demand at $0.87 an hour in US dollars as printed; the row is not on the console every hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.22 | patch | One more Sesterce row is read (first hourly read, 14:40 UTC). Its console lists an eight-GPU H200 machine with 1,128 GB, on demand at $35.20 an hour in US dollars as printed, which is $4.40 per GPU; the row is not on the console every hour. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.21 | patch | One more Sesterce row is read (first hourly read, 16:40 UTC). Its console now lists a GeForce RTX 4090 with 24 GB, one GPU on demand at $0.66 an hour, in US dollars as printed; its four-GPU machine costs the same per GPU. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.20 | patch | One more Sesterce row is read (first hourly read, 04:40 UTC). Its console now lists an RTX PRO 6000 Blackwell with 96 GB, one GPU on demand at $2.41 an hour, in US dollars as printed. The listing does not name the edition, so it is read as the RTX PRO 6000 Blackwell without an edition, as the plain rows of other providers are; the Max-Q card stays separate. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.19 | patch | Three more E2E Networks rows are read (first hourly read, 22:41 UTC). Its pricing page now prints the B200 at 180 GB, the card's usable memory, instead of 192 GB, and the row had dropped out at 10:40 UTC; it is read again, at $7.99 per GPU-hour (it had been $6.99). Two rows new on the same table are read too: the V100 32 GB at $1.50 and the T4 at $0.45 per GPU-hour, on demand, in US dollars as printed. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.18 | patch | One provider is read (first hourly read, 19:40 UTC). MimicPC, which hosts ComfyUI and other image apps on machines billed by the hour: the machine grid of its static pricing page, machines naming one card only, the whole machine with its RAM and workspace, in US dollars as printed - T4 $0.49 and L40S $1.99 an hour on demand (Instant Access), and $0.29 and $1.19 for its Bargain tier, which the page says can be interrupted and is listed as spot. Its machines listed as "L4•A10" are not read, since the page does not say which card a renter gets. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.17 | patch | One provider is read (first hourly read, 10:40 UTC). Bamboozle Web Services, a UAE cloud operator: the GPU instance tables of its static GPU compute page, the one-GPU flavor of each card, which includes vCPUs and RAM, in Dubai, in US dollars as printed - T4 $0.55, A100 80GB $2.50 and H100 $4.50 an hour. The page does not say whether its H100 is the SXM or PCIe edition, so it is listed as an H100 with the form factor not stated; its H200 is priced on request and is not read. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.16 | patch | Edgevana's A40 is read (first hourly read, 07:39 UTC): its card page sells a 4-GPU A40 machine on demand at $1.84 an hour, $0.46 a card, now a further source for the A40. It had been left out on the fleet index's figure, $2.09, more than three times the card's median, which the card page shows is not what the machine costs. No other figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.15 | patch | One provider is read (first hourly read, 19:40 UTC). Trooper.AI: the price table of its static prices page, one row per server type offered in several sizes, so each figure is the smallest size, one card, tagged as an advertised-from price and recorded in euros as printed - RTX A4000 EUR 0.21, V100 16 GB EUR 0.22, RTX 4070 Ti SUPER EUR 0.26, V100 32 GB EUR 0.28, RTX 3090 EUR 0.34, A100 40 GB EUR 0.61, RTX 4090 EUR 0.63, RTX PRO 4500 EUR 0.71, RTX 5090 EUR 1.13, RTX PRO 5000 EUR 1.30 and RTX PRO 6000 EUR 1.93 an hour. The lower promotional column beside them is the same rate paid with the bonus credit a deposit buys this month, not a separate price, and is not read; modified 48 GB RTX 4090 and 32 GB RTX 4080 SUPER boards and the RTX PRO 4000, whose edition the page does not name, are not read. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.14 | patch | One provider is read (first hourly read, 16:40 UTC). Barrack AI: the GPU Pricing card of its pricing page, which a script draws, so the page is rendered and read as a visitor sees it - one on-demand hourly rate per GPU in US dollars, as printed: B300 $8.99, B200 $6.89, H200 $4.45, H100 SXM $3.59, H100 PCIe $2.89, A100 80GB $1.49, L40 $1.19 and RTX A6000 $0.59. Its H100 PCIe on an NVLink bridge ($3.19) is the same 80 GB card; the cheaper row is the quote and names itself in the instance label. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.13 | patch | One provider is read (first hourly read, 10:41 UTC). HPC-AI.COM: the On-Demand Cloud GPU table of its pricing page, every rate printed "From" per card of a machine rented whole, eight cards at a time, so tagged as an advertised-from price and recorded per GPU as printed - B200 $4.00, H200 $2.68, H100 SXM $2.20 and RTX 5090 $0.62. Reserved clusters read "Contact Sales", the batch-job rate is a queued service and not read, and the RTX 4090 row is not read because it is printed with 40 GB, which no RTX 4090 has. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.12 | patch | One card is added (first hourly read, 22:40 UTC). The NVIDIA Quadro RTX 4000 (8 GB, Turing), on two sources 1.12x apart: Edgevana's card page sells one GPU on demand at $0.10 an hour, and the Vast.ai single-GPU floor under 10 GB is $0.0889 over exactly three verified asks. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.11 | patch | One provider is read (first hourly read, 16:39 UTC). Elice Cloud: the GPU table of its pricing page, drawn by a script and read as a browser draws it - the rate charged for the one-GPU instance of each card, never the struck-through list figure printed beside it, with vCPUs and memory included, so tagged as an instance price. On demand B200 $6.82, H100 SXM $3.57 and A100 80GB $1.62; spot $6.14, $3.21 and $1.46. The larger instances are the same rate times the GPU count; reserved is "Contact sales" and not read. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.10 | patch | One provider is read (first hourly read, 10:39 UTC). FPT AI Factory: the GPU Virtual Machine tab of its pricing page, the one-GPU machine of each card with CPU, memory, disk and a public IP included, so tagged as an instance price - B300 $6.99, H100 SXM $2.99, RTX PRO 6000 $2.19 and A30 $1.06 in Southeast Asia, H200 $6.60 in Japan. The larger machines on the page are the same figure times the GPU count; the container tab repeats two of these figures and is not read, nor are the managed inference tabs. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.9 | patch | Two cards are added (first hourly read, 22:40 UTC). GeForce RTX 3070 Ti (8 GB) on three sources: Edgevana one GPU on demand at $0.20, the Vast.ai single-GPU floor $0.1237 over three verified asks, and a Clore.ai 6-GPU machine at $0.9013 ($0.1502 a card). Quadro P4000 (8 GB) on two: an Edgevana 2-GPU machine at $0.14 ($0.07 a card) and the Vast.ai single-GPU floor $0.057 over three verified asks. Both Vast.ai books stand at the three asks a floor needs and read "thin market" on an hour they hold fewer. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.8 | patch | Two providers are read (first hourly read, 16:39 UTC). RunComfy: the pay-as-you-go column of its ComfyUI machine table, billed by the second, each figure the whole machine with one GPU and the vCPUs and memory beside it, so tagged as an instance price - RTX A6000 $2.50, A100 80 GB $4.99, H100 $7.49 (form factor not stated), H200 $9.59. Machines that name two possible cards, and the lower rates that need a monthly subscription, are not read. Phala Cloud: GPUs inside trusted execution environments; its GPU page draws the prices with a script, so it is read as a browser shows it - H100 $3.08 (form factor not stated) and H200 $4.80 on demand per GPU-hour, billed with a 24-hour minimum, and the B300 at $6.50 for one month, read as a reserved rate. A second price beside each, a reserved slot with no term printed, is not read. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.7 | patch | Three providers are read (first hourly read, 10:39 UTC). VESSL AI: the on-demand table of its Container Compute, a container per GPU billed per second, tagged as an instance price - H100 SXM $2.98, A100 80 GB $1.48, L40S $1.80. Hivenet: its pricing page draws the GPU table with a script, so it is read as a browser shows it; the one-GPU RTX 5090 instance (8 vCPUs, 73 GB of memory, 125 GB of disk) at EUR 0.75 an hour before tax, $0.8405 at the ECB rate. TensorPool: only its own card on a page that also shows other clouds' prices for comparison, the GPU alone since CPU and storage are charged apart - B200 $4.99, H200 $2.99, B300 $5.49, H100 SXM $1.99, L40S $1.49. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.6 | patch | One card is added (first hourly read, 04:39 UTC): the NVIDIA TITAN V (Volta, 12 GB), on two sources 1.03x apart. Edgevana's page for the card sells one GPU on demand at $0.14 an hour, and the Vast.ai single-GPU floor is $0.1356 over three verified offers, all 12 GB. The 32 GB CEO Edition is a different card and is kept out by memory on both sources. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.5 | patch | One provider is read (first hourly read, 22:39 UTC): Servercore, the Selectel group's international cloud, with its own site and price list. Its Cloud-Based GPU Servers page prints each price in five currencies; the US-dollar figure of the hourly column is read as printed. Each figure is a whole virtual machine with one GPU - the smallest configuration in its calculator, 8 vCPUs, 16 GB of memory and a small disk - so it is tagged as an instance price: RTX A5000 $1.12, A100 80 GB $6.82, RTX PRO 6000 $4.21, H200 $8.83, form factors not stated by the page. The A100 figure is now the highest rate for that card on this site. The page says nothing about VAT. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.4 | patch | Two providers are read (first hourly read, 16:39 UTC). Bitdeer AI (Singapore): its GPU pricing page draws its table from a price file it loads itself, which is what is read - per model, sizes of 1 to 8 cards (the GB200 NVL72 in trays of 4) with an on-demand rate and one- and three-year commitments, every figure printed "from" and so tagged "from". One quote per card and term at the size with the lowest price per GPU, the smaller size on a tie: H100 SXM $2.36 on demand, $2.0768 for one year and $1.7936 for three; H200 $2.761, $2.5125 and $2.264; B200 $9.38 for 2 cards ($4.69 a GPU; one card is $5.159), $8.7234 and $8.1606; GB200 $27.764 for 4 ($6.941 a GPU), $24.15468 and $19.99008. The page shows the commitment rates rounded to the cent. Krutrim Cloud (Ola Krutrim, India): the GPU virtual machine table of its compute documentation, on-demand rates in rupees for a whole VM - A100 80 GB 189 rupees ($1.95) and H100 213 rupees ($2.20), the H100 with its form factor not stated; the page's one- and six-month reserved rates and its AI Pods table are not read, and it says nothing about GST. No other provider's figure changed. The definition of the published figure did not change. | unchanged | |
| 1.33.3 | patch | Three providers are read (first hourly read, 10:39 UTC). Yandex Cloud: its Compute Cloud GPU price list bills the GPU apart from vCPU and memory, so the figure is the GPU line alone, in roubles with VAT as the list states - A100 80 GB 408.12 roubles ($4.81) on demand and 163.25 ($1.92) preemptible, V100 32 GB 261.28 ($3.08) and 65.34, T4 70.27 ($0.83) and 17.56. centron (Germany): five dedicated GPUs, the GPU alone, in euro net of VAT, the hourly figure as printed and rounded - RTX A4000 EUR 0.13, Quadro RTX 6000 EUR 0.24, A100 40 GB EUR 0.679, A100 80 GB EUR 1.243, RTX 6000 Ada EUR 1.192. YTL AI Cloud (Malaysia): the GB200 at 68.80 ringgit a GPU-hour ($16.84), storage priced apart. The definition of the published figure did not change. | unchanged | |
| 1.33.2 | patch | A new card, the NVIDIA Tesla P100 12GB (Pascal, PCIe), on two sources 1.14x apart: Edgevana's card page sells a 4-GPU machine at $0.43 an hour, $0.1075 a card, kept as the machine's price with its GPU count (first hourly read, 04:39 UTC), and the Vast.ai multi-GPU floor under 14 GB is a 4-GPU machine at $0.3778, $0.0944 a card (offer 53389245, first hourly read 05:39 UTC). The 16 GB P100 is another card and is kept out by memory. The definition of the published figure did not change. | unchanged | |
| 1.33.1 | patch | A new card, the NVIDIA RTX 4500 Ada Generation (24 GB), on two sources 1.15x apart: Edgevana's card page sells a 2-GPU machine at $0.93 an hour, $0.465 a card, kept as the machine's price with its GPU count, and the Vast.ai floor is $0.4047 over four free verified single-GPU offers, every one 24 GB (first hourly read, 22:39 UTC). The same read adds two Edgevana card pages for cards already listed: the RTX A2000 at $0.17 for one GPU, held out until now at the fleet index's $1.23, and the L40 at $1.05 for a 2-GPU machine, $0.525 a card. The definition of the published figure did not change. | unchanged | |
| 1.33.0 | minor | A new rule about which reads count as current: a quote whose provider's own page for the card answers "not found" in a read is unconfirmed from that read on, instead of live for the rest of the 48 hours, and live again once the page prints a figure. It leaves the site's computed figures at that read; the Price Index, computed in the database at each scrape, keeps its 48-hour rule. First source to report it: Edgevana's card pages. The definition of the published figure did not change. | unchanged | |
| 1.32.37 | patch | A new card, the NVIDIA A2 (16 GB), on two sources 1.32x apart: Infomaniak's cheapest A2 virtual machine at CHF 0.1623 an hour net of VAT ($0.1947) and immers.cloud's one-GPU A2 machine at 21.94 roubles ($0.2569) (first hourly read, 10:41 UTC). Both price lists carried the card before; it had no id here. The definition of the published figure did not change. | unchanged | |
| 1.32.36 | patch | Infomaniak (Switzerland) is read: its public-cloud price page loads the whole price list from a data file it names itself, and each card's quote is its cheapest GPU virtual machine, in Swiss francs net of VAT as the page prints it. First hourly read, 10:41 UTC: T4 CHF 0.2367 ($0.2839), L4 CHF 0.2824 ($0.3387), L40S CHF 0.7045 ($0.8450). Machines with two or four L4s are not read, because the source does not state their GPU count, nor the A100 machine, whose memory it does not state. The definition of the published figure did not change. | unchanged | |
| 1.32.35 | patch | Hetzner publishes one quote per card. Since 2026-10-06 its GPU matrix carries two servers with the same RTX PRO 6000 Blackwell Max-Q: GEX63 at $1.9214 an hour and GEX131 at $2.7227 "starting from". Both were written under one quote key, and the figure shown was whichever row the hourly run wrote last ($2.7227 on 2026-10-09). The parser now keeps the server cheapest per GPU, named in the row - the rule Edgevana and Sesterce follow - so the quote reads $1.9214 (GEX63). Not in the index; the definition of the published figure did not change. | unchanged | |
| 1.32.34 | patch | A new card, the TITAN Xp (12 GB), on two sources 1.24x apart. Edgevana's card page sells a 7-GPU machine at $0.54 an hour, $0.0771 a card, kept as the machine's price with its GPU count; the Vast.ai floor is $0.0622 over the free verified single-GPU offers, every one 12 GB, and its multi-GPU book prices a 7-card machine at $0.4689, $0.0670 a card (first hourly read, 05:41 UTC). Edgevana's row had been held out since 2026-10-05 at $0.37 as four to seven times the marketplace; that was the fleet index's figure, and the card page shows it for a whole machine. The definition of the published figure did not change. | unchanged | |
| 1.32.33 | patch | A new card, the TITAN RTX (24 GB), on two sources 1.14x apart. Edgevana's card page sells a 2-GPU machine at $0.55 an hour, $0.275 a card, kept as the machine's price with its GPU count; the Vast.ai floor is $0.2422 over the free verified single-GPU offers, every one 24 GB, while its multi-GPU book holds two machines and reads "thin market" (first hourly read, 23:41 UTC). Until this hour the Vast.ai book held fewer than the three verified offers a floor needs, so the card had a single source and no id here. The definition of the published figure did not change. | unchanged | |
| 1.32.32 | patch | Edgevana's RTX 5060 is read from its card page, a fourth source for the card and its second rate card: one GPU on demand at $0.11 an hour, beside the Vast.ai floor of $0.0994 for the same 8 GB card (first hourly read, 22:41 UTC). The row had been held out since 2026-10-05 at $0.39 as more than three times the card's median; that figure was the fleet index's cheapest configuration, not this card page's one-GPU rate. The definition of the published figure did not change. | unchanged | |
| 1.32.31 | patch | A new provider, Hikube (Hidora SA, three Swiss data centres), read from the public catalogue on its pricing page. It prices six GPUs by the month, in Swiss francs and excluding VAT, and bills by the hour; the page states its own divisor, "Hourly = monthly ÷ 730 h", so each rate is its monthly price over 730 hours, which the owner accepted on 2026-10-08 for a page that states the divisor - the live check that a figure is printed on the linked page takes the month over 730 for this provider only. Each figure is the GPU alone, added to an instance priced separately. First hourly read, 16:41 UTC, at the ECB rate of 2026-10-08 (1.199442): L4 CHF 165 a month, 0.2260 an hour, $0.2711; L40S 760, $1.2487; A100 80 GB 1495, $2.4563; RTX PRO 6000 1550, $2.5468; H100 1999, $3.2846 (the page names no form factor, so it is the form-factor-neutral H100 and not the index card); H200 2499, $4.1061. The definition of the published figure did not change. | unchanged | |
| 1.32.30 | patch | A new provider, IONOS Cloud, read from its official euro price list (docs.ionos.com, version of 2026-09-15). The list prints four Cloud GPU VM templates of one card, the H200 in its PCIe form - the card NVIDIA sells as the H200 NVL - with 1, 2, 4 and 8 GPUs at EUR 3.00, 6.00, 12.00 and 24.00 an hour, exact multiples, so the one-GPU template H200-S is the quote: EUR 3.00, $3.3531 at the ECB rate of 2026-10-07 (first hourly read, 10:41 UTC), with its 15 dedicated CPUs, 267 GiB of memory and 1 TB NVMe disk. The list says its prices exclude VAT, and the figure is kept as listed. The PCIe card never enters the SXM H200 or the index. The definition of the published figure did not change. | unchanged | |
| 1.32.29 | patch | Sesterce is read again, from a different page. Its home page carried a "From / GPU / hr" table rendered from its console's live offers; on 2026-10-08 the home page was redesigned, the table went, and the 04:37 and 05:37 UTC runs recorded the source as failing rather than quiet. The console itself, cloud.sesterce.com/compute, now prints the same offers in the page it serves, with no account: "Launch On-Demand VMs & Bare Metal Servers. No commit.", one row per card and GPU count, priced for the whole machine. That page is what each row links to now. One quote per card, the count that is cheapest per GPU, kept as the machine's price with its GPU count, and still a "from" figure (the row is the cheapest of the offers behind it). Fifteen cards in the read this change was checked against (06:21 UTC), among them H100 $2.75 for one GPU (the console names no form factor, so it is the form-factor-neutral H100), A100 80 GB $2.97 for two, L40S $3.30 for two, RTX 5090 $0.71. The console lists no B200, B300 or H200 today, so the three quotes the home page carried for them stop with the time each was last confirmed. Each row's memory is checked against the card its name maps to; a mismatch is reported, never mapped. No figure of any other provider changed. The definition of the published figure did not change. | unchanged | |
| 1.32.28 | patch | A new card, the GTX 1070 Ti, on two sources 1.16x apart. Edgevana's card page sells a 6-GPU machine at $0.42 an hour, $0.07 a card, kept as the machine's price with its GPU count; the Vast.ai multi-GPU book has a 4-card machine at $0.2422, $0.0606 a card, while its single-GPU book holds one verified offer and reads "thin market" (first hourly read, 05:41 UTC). Edgevana's row had been held out since 2026-10-05 as four to seven times the marketplace: the $0.42 was always a whole machine, which only the card page states. The definition of the published figure did not change. | unchanged | |
| 1.32.27 | patch | The RTX 4070 SUPER, on this site since 2026-10-07 with the Vast.ai floor alone, has its second source: Edgevana's card page sells an 8-GPU machine at $1.12 an hour, $0.14 a card, kept as the machine's price with its GPU count (first hourly reads 03:41 and 04:40 UTC), beside the Vast.ai floor of $0.2202. The card was off Edgevana's index for the first hourly read after it was mapped, which is why this entry comes a day after the card. The definition of the published figure did not change. | unchanged | |
| 1.32.26 | patch | A new card, the RTX 2060 12GB, on two sources 1.03x apart. Edgevana's card page sells a 2-GPU machine at $0.22 an hour, $0.11 a card, kept as the machine's price with its GPU count; the Vast.ai floor is $0.1071 over the free verified single-GPU offers, every one 12 GB, with a 2-card machine at $0.2138 (first hourly read, 23:41 UTC). The RTX 2060 also shipped with 6 GB, which is another card with no id here: Edgevana's mapping is pinned to the 12 GB its row prints and Vast.ai is read from 10 GB up. The definition of the published figure did not change. | unchanged | |
| 1.32.25 | patch | Edgevana is read from each card's own page, no longer from its fleet index. The index prints one hourly figure per card, the cheapest available configuration of that card, whatever it is - and when the single-GPU configurations were sold out that was a whole multi-GPU node, or a reserved term's rate, published here as one GPU's on-demand price: the H200 at $55.70 against a median of $5.89, the H100 PCIe at $24.72. Edgevana's own card pages list every configuration with its SKU, the GPU count it states ("2x H200 NVL"), an hourly figure for the whole machine and the per-GPU rate printed beside it, and say whether it is on-demand or reserved. A card is now published only from a configuration that is on-demand, names the card itself, has the memory the mapping is pinned to, states its GPU count and is available; the cheapest per GPU is kept, a machine priced whole is recorded per node with its count, the instance is the configuration's own SKU, the source link is the card's page and the scope is instance, no longer "advertised from". A card whose page is not served - the index links it, the page answers 404 - or has no such configuration is not published and is named in the run's report; the index's figure is never a fallback. The run now makes one read of the index and one of each card page, about forty an hour. No provider was added or dropped and the definition of the published figure did not change; the figures Edgevana now publishes are the ones the page states for a configuration, and some cards it carried on the index (H100 SXM, H100 PCIe, H100 NVL, H200 NVL, GH200, RTX 3090 Ti, RTX 6000 Ada, RTX PRO 6000) have no card page today and are not published until they have one. | unchanged | |
| 1.32.24 | patch | Lyceum is read again, and the first provider this site has lost to a redesign came back with a price column it did not have before. Its pricing page had carried three GPU tables since it was first read on 2026-09-26, and the reader keyed on a blurb over one of them, "Full root access". The page was rebuilt, the three tables went, and from 06:40 UTC on 2026-10-05 every hourly run recorded the source as failing rather than quiet - which is the right outcome for a scraper whose page has changed, and it is why nothing wrong was published for two days: the six rates went to unconfirmed and then stopped, with the time each was last confirmed beside it, and the provider left the panel. It was read again at 09:41 UTC on 2026-10-07, fifty hours later. What stands on the page now is one table, headed "GPU compute", and its own caption states the unit - "USD per GPU hour, billed per second with no subscription" - so that sentence, and not a product blurb, is what the reader now keys on. The six on-demand figures came back to the cent as the old table printed them: L40S 48 GB $1.19, A100 80 GB $1.59, H100 $2.79, H200 141 GB $4.29, B200 192 GB $6.49, B300 $7.99. Each sits inside the corridor its card's other sources make - the furthest from a median is the H100 at $2.79 against $3.47 over twenty live quotes - so the rule that holds back a figure more than three times from a card's other readings holds back none of them. The table's second priced column, Spot, is new on this page and is read too, as five more quotes: A100 80 GB $0.80, H100 $1.10, H200 $1.60, B200 $2.40, B300 $2.99. The L40S row says "Not offered" where its spot price would be, and no figure is invented for it. Each row prints its two prices twice, once as the labels of a bar chart in the bars' own order - the B300 row reads "Spot $2.99" before "On demand $7.99" - and once as labelled cells; only the labelled pair is read, so the cheaper bar never becomes the on-demand rate. Three things on that page are read and not published, because none of them is an hourly rate for a card: the list of per-token model prices above the table, the three products below it (dedicated endpoints, serverless training, reserved GPUs and clusters), each of which answers the question of price with "Commitment pricing, agreed per contract", and the Reserved blocks column, which offers capacity - "Reserved 64 · Europe or Middle East · from Q4 2026" - with no figure beside it. This entry also records a source that has been read since 2026-10-05 and was never written up here, because the run that added it ended before it could be: QuantaCloud, quantacloud.net/pricing, a catalogue of launchable configurations whose every figure stands in the served page. Seven cards, one quote each, read first at 10:10 UTC on 2026-10-05: RTX A6000 $0.48, RTX 6000 Ada $0.79, L40 $0.94, L40S $1.09, A100 PCIe 80 GB $1.48, RTX PRO 6000 Blackwell $2.39, H100 PCIe $2.59, every one inside its card's corridor. The catalogue sells the same card at several node and host sizes and the rate per GPU is not the same across them - the RTX A6000 is $0.48 a GPU in four configurations and $0.54 and $0.56 in two others - so the published figure is the cheapest single-card configuration and the configuration's own partition travels with it, which is the rule two other catalogues here are read under. Its form factors are read and not guessed: the catalogue writes "H100 PCIe" and "A100 SXM4 80GB" where its own summary table says only "H100". Seventy-five providers now carry a live quote, which is every source this site reads. No figure of any other provider or card changed, and no rate already published moved. The definition of the published figure did not change. | unchanged | |
| 1.32.23 | patch | Two cards join the catalogue in the same hour, the NVIDIA GeForce GTX 1080 Ti (11 GB) and the NVIDIA GeForce GTX 1080 (8 GB), and the first of them is the first card on this site to arrive with three sources rather than two. All three were already read every hour. Edgevana's fleet index prints the 1080 Ti at $0.07 and the 1080 at $0.06; SynpixCloud's market page prices the 1080 Ti at $0.23 for a KVM instance of 8 vCPU, 24 GB RAM and 200 GB disk that the page marks Available; and the Vast.ai marketplace gives each card a trimmed floor over the free verified single-GPU asks in its book - $0.0649 for the 1080 Ti over five asks from $0.0422 to $0.0649, every offer 11264 MB, and $0.0529 for the 1080 over five from $0.0529 to $0.0556, every offer 8192 MB. Each card also carries a second quote from that marketplace's multi-GPU partition: a two-card machine at $0.1276 an hour, or $0.0638 a card, for the 1080 Ti, and a four-card machine at $0.2049, or $0.0512 a card, for the 1080. 11 GB and 8 GB are the only sizes NVIDIA shipped these cards in, so neither book needs a memory window, and the name is the whole guard between the two: the marketplace is asked for an exact name, so a request for the 1080 never draws the Ti's offers. The 1080's price list and floor are 1.13x apart, which is the closest two sources have been on the day a card joined this site. Where the 1080 Ti's three readings spread wider - $0.23 against $0.0649 is 3.5x - the rule that holds back a figure more than three times from a card's other readings does not apply, and the reason is what that rule is for: Synpix's $0.23 is its own cheapest tier, the same cell its RTX 2080 Ti and RTX 3080 stand in, and a tier that prices three unrelated cards alike is a price list, not a placeholder sitting in cells that have nothing to do with one another. All three names had been read and dropped as unmapped from the first hour each source was read, for want of an id and nothing else; Synpix's own note said so in as many words - "no row in gpus yet" - which is a threshold and not a refusal. Eleven further cards were looked at in the same read of the marketplace and are not listed. Four of them have a figure on both sources and the two disagree too far to publish: Edgevana prices the Titan Xp at $0.37 against a floor of $0.0622, the Tesla P100 at $0.43 against $0.0944 a card, the GTX 1660 SUPER at $0.28 against $0.0582 and the GTX 1070 Ti at $0.42 against $0.0606 a card. Seven have one source and no second. A third source also joined a card that arrived yesterday: Clore.ai's RTX 3060 Ti is now read, and its multi-GPU partition publishes a four-card machine at $0.3063 an hour, $0.0766 a card, between Edgevana's $0.11 and the Vast.ai floor's $0.0941, while its single-GPU book held one free listing where a floor needs three and therefore reads "thin market". A row on Edgevana's index naming the B300 at $138.01 an hour is read and not published: that is fourteen times the dearest reading this site has of that card, from DataCrunch at $9.8175 and Oracle at $18.00, and a figure that far from a card's other readings is held back until a reading explains it. No figure of any other card or provider changed, and no rate already published moved. The definition of the published figure did not change. | unchanged | |
| 1.32.22 | patch | Two cards join the catalogue in the same hour, the NVIDIA GeForce RTX 3060 Ti (8 GB) and the NVIDIA GeForce RTX 4060 (8 GB), each priced by two sources that were already read every hour. Edgevana's fleet index prints the 3060 Ti at $0.07 an hour and the 4060 at $0.16, and the Vast.ai marketplace gives each a trimmed floor over the free verified single-GPU asks in its book: $0.0948 for the 3060 Ti, over five asks from $0.0622 to $0.0956, and $0.0815 for the 4060, over four from $0.0763 to $0.1356 - every offer in the first book 8192 MB and in the second 8188 MB, which is each card's only size, so neither needs a memory window and neither can take in the laptop part of the same number. The 3060 Ti carries a second quote from the marketplace's multi-GPU partition, a four-card machine at $0.6437 an hour or $0.1609 a card; the 4060's multi-GPU book held one free machine where a floor needs three, so that quote is withheld and reads "thin market" until the book turns over. Both names had been read and dropped as unmapped by both sources from their first hour, for want of an id and nothing else. What changed is which sources stand beside them: until 2026-10-04 the only second source either card could have had was the other marketplace, Clore.ai, and this site does not add a card that stands on two marketplace floors alone, because a floor moves with supply and is not a rate card. Edgevana, read since 2026-10-04, is a published rate card and prices both, so each card arrives in the ordinary shape - one price list and one marketplace floor - that nine consumer cards here arrived in before them. Clore.ai spells both names as well and published neither this hour: two whole machines under the 3060 Ti and no free single-GPU one, where a floor needs three of either. A third card was read in the same pass and is not listed: the RTX 4070 SUPER has two readable sources, but Edgevana's $0.99 for it is five times its own RTX 4070 and RTX 4080 on the same page and 7.3 times the marketplace floor for the same card, and a figure that far from a card's other readings is held back until a reading explains it. No figure of any other card or provider changed, and no rate already published moved. The definition of the published figure did not change. | unchanged | |
| 1.32.21 | patch | Edgevana's L40S row is read, so the NVIDIA L40S is priced by a thirty-fifth provider and carries a forty-eighth live quote. The name had stood in that page's "Standard - 48-80GB" table from the first read on 2026-10-04 and the parser's allowlist had never been given it, so it was reported and dropped on every hourly read; mapping it adds a rate from a page already read every hour and asks nothing new of the source. The row prints 45 GB, which is the driver figure for a 48 GB card and the same one the A40 row prints beside it, and the mapping is pinned to that 45 GB: a row that stops printing it is reported rather than read, as every mapping on this source is. The figure moved between the two readings of the day and both were kept as read: $0.80 an hour at 16:04 UTC, two configurations in two regions marked Limited, and $1.84 when the hourly run published it at 16:40 - a 2.3-fold move on one row inside 36 minutes, wider than anything this catalogue had shown before, and the published figure is the one the hourly read took. Both numbers sit inside the corridor this card's other sources make (the 47 live quotes it carried before this one, from $0.57 to $3.51 with a median of $1.20), so neither is a placeholder and the rule that holds back a figure more than three times from a card's median holds back neither; what the pair shows is that a rate from this source is a reading of its minute, which is why every figure here is tagged as the "from" rate the page prints and carries the time it was read. The same rule did hold one row back in the same reading: Edgevana's RTX 5060 at $0.39 against that card's median of $0.095 over seven live quotes is 4.1 times the median and 2.5 times the dearest quote this site has, so it is reported and not published, beside the A40 and RTX A2000 rows already held back there. One card did not join the catalogue today and was briefly listed by mistake: the NVIDIA RTX PRO 4000 Blackwell was mapped from two sources at 16:40 UTC and both quotes were retired and the mapping reverted within the hour, because this site does not list that card while no source names its edition - NVIDIA ships a full-size board and an SFF board both at 24 GB, and neither source names which it sells. Its two quotes reached no page beyond the hour they were written, and no figure of any other provider or card changed. The definition of the published figure did not change. | unchanged | |
| 1.32.20 | patch | Edgevana is read, and four of the seven cards that stood on fewer than three sources now stand on three: the RTX PRO 5000 Blackwell at $0.69, the RTX 4060 Ti at $0.16, the RTX 3060 at $0.07 and the RTX 5000 Ada at $0.39. nodes.edgevana.com/gpu-servers is one index of the whole fleet in plain HTML - three tables banded by memory, one row per GPU model, columns GPU Model, VRAM, Configs, Hourly, Monthly, Locations, Availability - and the page says what it sells in its own words: "72 GPU models, 489+ configurations across 15+ global locations. Bare metal, single-tenant, instantly deployable." Every rate from it is scope `advertised_from`, because a row is not a machine: it is a card with N configurations behind it in M regions and a single hourly figure, that is, the cheapest of those N. Nothing on the page states what one configuration contains, so the figure is published as the "from" rate it is printed as and never as an assembled instance, and no region is recorded - the page prints how many locations stand behind a card, never which. The provider's own name for each row travels with the quote, so a reader opening the link finds the row the figure came from. The Monthly column is not read: it is the same capacity sold by the month, and dividing it would be arithmetic of this site's rather than a figure the page prints. Four kinds of row on that page are reported and deliberately not published. Whole nodes under a card's name - NVIDIA B200 at 1536 GB for $81.60 an hour and NVIDIA H100 at 640 GB for $28.70 - are eight-card machines, and dividing their memory by a card's to get a GPU count is an inference this site does not make. A second naming family that does not agree with the first - GPU-H100-SXM at 80 GB for $43.80 standing beside the plain H100 SXM at 80 GB for $2.18, and GPU-H200-SXM and GPU-B200-SXM likewise - is left alone because $43.80 for one H100 is fifteen times this site's median and nothing on the page settles what those rows price. Capacity the page marks Pre-order is not a rate for anything that can be rented today, and both such rows also say 0 regions. And four rows fall more than three times from this site's median for their card, in either direction - the A40 at $2.09 against a median of $0.56, the RTX A2000 at $1.23 against $0.1276, and both A100 40 GB rows at $0.39 and $0.62 against a median of $1.98 and a market minimum of $0.4157 - so they are held back until the per-card page can settle them, the same corridor rule that caught RunPod's $0.50 placeholder. Where two rows name one card, one quote is published at the cheaper of them and never two: RTX PRO 6000 S at $1.68 and RTX PRO 6000 WS at $1.51 are both 96 GB and both the RTX PRO 6000 Blackwell, and two quotes would share a key and count this provider twice. This catalogue moves inside the hour, which is why each mapping is pinned to the memory its row prints: between 10:04 and 10:26 UTC the H100 SXM went from $2.18 to $2.24, the L4 from $0.77 to $0.37, and A100 PCIE changed its stated memory from 40 GB to 80 GB, which is two different cards under one name. A row that stops matching the memory it was mapped at is therefore reported rather than read, and the first hourly run published 36 rates where the check twenty minutes earlier saw 39, the three absent ones each a single-configuration row the page marked Limited. No figure of any other provider or card changed, and no card arrived or left the catalogue. The definition of the published figure did not change. | unchanged | |
| 1.32.19 | patch | Hugging Face is read in two books instead of one, so the NVIDIA A10G (24 GB) stands on three sources instead of the two it has had since September 27. huggingface.co/pricing carries two price lists, and this site read one of them: the Spaces Hardware table was skipped on the ground, written into the parser since September 26, that it "repeats these figures for Spaces". Read off the live page at 22:14 UTC on 2026-10-04, that ground is wrong in two places. The A10G is priced in Spaces Hardware - $1.00 an hour for `Nvidia A10G - small`, $1.50 for `- large` - and appears in no Inference Endpoints row at all, that section's eight GPU rows being the T4, L4, L40S, A100, H100, H200, B200 and RTX PRO 6000; and the T4 is $0.40 in Spaces against $0.50 under Inference Endpoints. A table that prices a card the other table does not price, at figures the other table does not carry, is not a repeat of it, so it is read as a second book of the same source under the region `Spaces hardware`, the way Dataoorts' four series and Clore.ai's multi-GPU partition are read. Where the two books do carry one figure - the L4 at $0.80, the L40S at $1.80, the A100 at $2.50 - that is the repeat the old ground described, and it is skipped as one figure twice rather than published as a second rate: the reading this site already applies to RunPod's spot rows, Vultr's preemptible plans and GPU.ai's two tiers. So the second book publishes exactly two quotes. The A10G's $1.00 is the first rate under a dollar this card has on the site, beside AWS's g5.xlarge at $1.006 and Baseten's $1.2072, and the T4 gains a second Hugging Face quote at $0.40, which the T4's median counts as it counts any provider's second partition. Read: the single-card rows only (a name with no multiplier, or `1x`), the cheapest row per card, because the sizes of one card there differ by the host CPU and memory and not by the card, and the 2x, 4x and 8x rows are multiples as the larger Inference Endpoints sizes are. The card is taken from each row's own Name column and never from its Accelerator column, which this page fills with `Nvidia L4` on all three of its L40S rows; `ZeroGPU` prints FREE and `Custom` prints "on demand", and neither is a rate. The eight Inference Endpoints rows are read exactly as before, and no figure of any other provider or card changed. The definition of the published figure did not change. | unchanged | |
| 1.32.18 | patch | Vast.ai's RTX 4070S Ti is read, so the NVIDIA GeForce RTX 4070 Ti SUPER (16 GB) stands on three sources instead of the two it arrived on. The marketplace had been selling the card all along under a name this site had never asked it for: it puts the SUPER's "S" on the model number rather than at the end of the name - the spelling it already uses for `RTX 4080S` - so `RTX 4070 Ti SUPER` and `RTX 4070 Ti S` match no offer here and were probed and dropped. The floor is $0.1923 (offer 53719523), the median of the five cheapest free verified single-GPU asks at 17:40 UTC on 2026-10-03, the same trimmed floor every Vast.ai figure is; it stands beside Clore.ai's floor $0.1969 and Salad's rate card at $0.2100 at High priority, the pair the card joined on on 2026-10-01. Read 80 minutes earlier the same window ran $0.1756 to $0.2156, every offer in it reporting 16376 MB, with the floor one offer along at $0.1963 - an hour of a marketplace book, not a disagreement, and the figure published is the one the hourly read took. The marketplace's multi-GPU partition is a second quote on this source, a machine of eight at $1.7081 an hour, $0.2135 a card, chosen out of its book by the per-GPU ask. The card is read at 14 GB and above, because the plain RTX 4070 Ti is a 12 GB card with its own id here and memory is what tells the two apart, not the name; on this source the bound travels inside the query, so a 12 GB machine is never served. Four cards that had no name on this marketplace were asked for in the same read and it carries none of them - the A10G, the RTX 2000 Ada, the RTX 4000 SFF Ada and the MI350X returned zero verified offers on every spelling tried - so nothing of theirs changes either. No figure of any other provider or card changed. The definition of the published figure did not change. | unchanged | |
| 1.32.17 | patch | The NVIDIA GeForce RTX 4060 Ti (16 GB) joins the catalogue on two sources, each checkable where its row links: Salad's rate card at $0.14 an hour at High priority ($0.12, $0.10 and $0.08 down its three preemptible tiers, its table printing 16 GB beside the name) and the Vast.ai floor $0.1148, the median of the five cheapest free verified single-GPU asks in its window at 10:20 UTC on 2026-10-03, $0.1030 to $0.1356, every one of them reporting 16380 MB - the same trimmed floor every Vast.ai figure is. The marketplace's multi-GPU partition is a second quote on that source, a machine of two at $0.2281 an hour, $0.1141 a card, chosen out of ten by its per-GPU ask. The two sources are 1.22x apart. The card is read with a memory window at 12 GB and above, because the 8 GB edition is sold under the same name and has no id here - the rule applied to the RTX 5060 Ti, "A100 SXM4" and "RTX 4090"; on this source the bound travels in the query, so an 8 GB machine is never served. Clore.ai spells the name the same way and is read for it in the same change, and publishes nothing: in its own 16 GB window one single-GPU machine was free beside two free whole machines, where a floor needs three of either, so both its books are withheld and that quote reads "thin market" until the book turns over - the card's third source arrives on its own the hour it does. No figure of any other provider or card changed. The definition of the published figure did not change. | unchanged | |
| 1.32.16 | patch | Five names Vast.ai has carried all along are read, so four cards that stood on two sources now stand on three: the RTX 4070 floor $0.1378 over 3 verified single-GPU offers, the RTX 5070 $0.2415 over 4, the RTX 5060 Ti $0.1348 over 5 and the RTX 5060 $0.0994 over 5 - each the median of the five cheapest verified, rentable offers at 04:12 UTC on 2026-10-03, the same trimmed floor every Vast.ai figure is. The RTX 2080 Ti is read too and publishes nothing: one verified offer carried the name, where a floor needs three, so its quote is withheld and reads "thin market" until the book fills. Two of the five take a memory window, because the marketplace sells two cards under one name: the RTX 2080 Ti is read below 16 GB, which keeps out the modified 22 GB part that has no id here, and the RTX 5060 Ti at 12 GB and above, which keeps out the 8 GB edition - the rule this site has applied to "A100 SXM4" and "RTX 4090" since 2026-09-17. Every offer in each of the four published books reports one memory size (12282, 12227, 16311 and 8151 MB), which is also what says these are the desktop cards and not the laptop parts of the same number. No figure of any other provider or card changed. The definition of the published figure did not change. | unchanged | |
| 1.32.15 | patch | The NVIDIA RTX 2000 Ada (16 GB) joins the catalogue on two sources, each checkable where its row links: RunPod's API (`NVIDIA RTX 2000 Ada Generation`, 16 GB) $0.24 on Secure Cloud, and GPU.ai's rate table (`RTX 2000 ADA`, 16G) $0.24 - both on 2026-10-03. RunPod's two community cells for the card are exactly the $0.50 the API returns in a dozen unrelated cells, above the card's own secure price, and are not published. It is the Ada-generation workstation card, not the RTX A2000 (6 GB, Ampere) already listed. The definition of the published figure did not change. | unchanged | |
| 1.32.14 | patch | GPU.ai's RTX 4000 Ada is read: its rate table prints `RTX 4000 ADA` at 20G and $0.28 an hour on 2026-10-03 (22:04 UTC), a name the module read and dropped until today. The RTX 4000 Ada and the RTX 4000 SFF Ada both carry 20 GB, so memory cannot separate them and the name does - this row prints no SFF, so it is the full-height card, as RunPod, Koyeb, Linode and Vast.ai already read the same pair of names. The figure stands inside the card's $0.20-$0.76 live range and, like every GPU.ai rate, is the lowest live on-demand rate per GPU the table prints. The definition of the published figure did not change. | unchanged | |
| 1.32.13 | patch | Hostinger is read: seven products on www.hostinger.com/gpu-hosting, each priced by the hour in the served HTML - B200 (Dedicated) $7.08, B200 $4.50, A100 80GB $1.55, L40S $0.96, RTX PRO 6000 $0.65, RTX 5090 $0.64 and RTX 4090 $0.42 on 2026-10-02. Every card prints "From" over its rate and the page's own FAQ says why - availability varies by region and changes in real time - so all seven are tagged as advertised rates rather than configured instances. Each rate is checked against the credits figure printed beside it, a hundred credits to the dollar in all seven rows, and the memory each card states is checked against the card it is mapped to, so a familiar name carrying a different memory size is dropped instead of published. The B200 appears twice because Hostinger sells it twice, with two names and two product ids behind the page's own buttons; only the dearer one says it is dedicated and non-shared, so that word is recorded as its partition and the cheaper row is left without one, since the page does not say what it shares. Both state the full 192 GB, so neither is a slice of the other. Its RTX PRO 6000 at $0.65 is the cheapest rate on the site for that card, one cent under Packet.ai's $0.66 for the same card. No figure of any other provider changed. The definition of the published figure did not change. | unchanged | |
| 1.32.12 | patch | Fireworks AI is read, and it is the third source for the GB300, which until today stood on DataCrunch and Oracle alone. Its pricing page carries several price lists and one of them prices a GPU by the hour: the table under its own "On-Demand Pricing" heading, where it bills per GPU second. Five cards on 2026-10-02 - H100 80 GB $8.00, H200 141 GB $8.00, B200 180 GB $13.00, B300 288 GB $15.00, GB300 288 GB $20.00 - each read from the per-hour column the page prints, not from the per-minute column beside it, which is the same rate rounded ($0.250 x 60 is exactly $15.00, while $0.134 x 60 is $8.04 against a printed $8.00). The two columns now cross-check each other: a row whose minute figure falls more than 2% from its hour figure stops the read rather than publishing a misread rate. The H100 and H200 share a rate, which is this provider's own choice and not a placeholder - each row's two columns agree and the three Blackwell rows below differ from one another. The per-million-token lists above the table are not GPU rates and are not read, and the Dedicated Training API section names no figure of its own, pointing at these same five rates. The figures are per GPU with no region: the page's note that region-restricted deployments cost 1.5x is a multiplication of ours and is not published. No figure of any other card or provider changed. The definition of the published figure did not change. | unchanged | |
| 1.32.11 | patch | The Price Index on this site's pages now takes one quote per provider and region, as its written method and its weekly receipts always did. GPU.ai, read since 2026-10-01, lists two H100 rates with no region label - an on-demand rate and a community rate - and the pages counted both while the receipts, computed in the database, counted one: the week of 2026-09-28 read $3.49 on 31 quotes on the page and $3.45 on 30 in its receipt. A figure that cannot be recomputed from its own receipt breaks the one thing the index promises, so the pages now keep, per provider and region, the newest read and, between two reads of the same instant, the last row written - the database's rule. Only the running week's figure on the page moves, to the receipt's; closed weeks, the hourly series and the monthly averages were computed in the database and are unchanged. The definition of the published figure did not change. | unchanged | |
| 1.32.10 | patch | The RTX 2080 Ti joins the catalogue, and on a shape new to this site: two published rate cards and no marketplace floor. SynpixCloud prices it at $0.23 an hour for a single-GPU machine its own Status column calls Available, and immers.cloud at 45.95 roubles an hour for a machine of two - 22.975 roubles a card, $0.276 at the Bank of Russia rate of 0.012013 for 2026-10-02 - which is that source's cheapest rate per card, against 25.05 roubles for its single machine. The two figures are 1.20x apart. It is the oldest consumer card on the site, a 2018 Turing part, and the one neither marketplace could carry: Clore.ai had a single unverified machine under the name where a floor needs three verified ones, and Vast.ai has no model for the name at all. Both sources state the 11 GB that is the whole of this card's identity - immers in its section heading, SynpixCloud in a VRAM column its own guard checks against the card's memory times the machine's GPU count, which is what keeps that page's 48 GB "RTX 4090PLUS" out of the RTX 4090's book. No figure of any other card or provider changed, and the card was read and dropped by both sources until today, so nothing published before now was wrong. The definition of the published figure did not change. | unchanged | |
| 1.32.9 | patch | The RTX 5060 joins the catalogue, on one published rate card and one marketplace floor - the shape three cards before it arrived in. Salad prices it at $0.11 an hour at High priority ($0.095 / $0.08 / $0.065 down its preemptible tiers), and the Clore.ai floor is $0.1794 over the five free verified single-GPU listings in its window at 04:13 UTC, the cheapest five asks $0.1728 to $0.9625; a second quote is the marketplace's multi-GPU partition, a machine of eight at $0.8711 an hour. That $0.9625 ask is why this card waited: it is an outlier five times the others, and the figure published is the median of the window rather than any average, so the outlier is absorbed and never becomes the price. Unlike the RTX 5060 Ti, which arrived on 2026-10-01 and whose one name covers an 8 GB and a 16 GB card, this name covers one card - all 17 machines carrying it report the 8 GB that is the only size NVIDIA ships the 5060 in - so its memory window divides nothing and only guards the name. The card was ready on 2026-10-01 and held for one day because Clore.ai answered every hourly run with a 404 from 18:37 UTC that day, which would have brought the card to the site on a single source; that source has answered three of the last four runs. The RTX 4060 Ti was read the same hour and is not published: Salad prices it, but the marketplace had one free verified listing inside its 16 GB window where a floor needs three. No figure of any other card or provider changed. The definition of the published figure did not change. | unchanged | |
| 1.32.8 | patch | fal's B200 rate is read again, and it had moved. fal rewrote that row's memory column from 180GB to 192GB between the 16:40 and the 17:40 UTC read on 2026-10-01, and because the memory is part of the name this site matches on, the row stopped matching: nothing wrong was published - the name simply went unread, which is what a key that no longer matches should do - but the figure stopped being confirmed, and the rate behind it had changed. The page prints $7.99 an hour where the last confirmed reading was $6.25, so for five hours this site showed a figure its source no longer carried, marked with the 16:40 UTC time it was last confirmed. Both spellings of the memory are now read, the new one being the more exact: 192 GB is what this site's own B200 record carries. At $7.99 the rate sits inside the pack it belongs to, between Together AI's $8.19 and DataCrunch's $6.99. The GB200 row the same table now prints is deliberately not read: this site's GB200 is the NVL72 rack at 186 GB, that row says 192, and no page fal links says which of the two the rate is for. The definition of the published figure did not change. | unchanged | |
| 1.32.7 | patch | Clore.ai's "H100 NVL" is read as the H100 NVL. It is the one datacenter name the sweep of that feed on 2026-09-30 left out, and it was left out for a reason that does not hold: that its book was a single machine, below the three a floor needs. A marketplace book turns over every hour, which is exactly why the six names mapped that day were mapped while none of them could carry a floor - read and withheld, a quote reads "thin market" and becomes a figure the hour three machines come free, where an unmapped name would stay silent. The mapping is the one Vast.ai's module already makes of the same name, so the two marketplaces cannot read one name two ways, with the memory window this module takes no name without: 94 GB is the only size NVIDIA ships the NVL in, so neither an 80 GB H100 nor a 141 GB H200 could enter the book. Nothing is published from it yet - at 22:09 UTC no verified and unrented single-GPU machine carried the name and the multi-GPU book held two where a floor needs three, so both quotes are withheld. The card already has eleven sources and no figure of any of them changed. The definition of the published figure did not change. | unchanged | |
| 1.32.6 | patch | Geodd is read: seven rates off its own pricing page, which carries its GPU cluster catalogue in the page as JSON - H200 NVL $3.39, H100 SXM5 $3.00, RTX PRO 6000 Blackwell $2.14, A100 SXM 80GB $1.49, L40S $1.00, Tesla V100 32GB $0.16, and the H200 SXM5 at $21.76 for a machine of eight. Each rate is a configuration with its own vCPUs, memory and disk, so each is the price of an instance, and only the smallest configuration of each card is published: 2x, 4x and 8x are exact multiples of the single-card rate, and four copies of one figure would weigh four times as much in the figures computed from them. The same rule settles the one place the catalogue contradicts itself - two 8-GPU H100 entries, $24.00 and $25.00 - without this site choosing between them, because the H100 is sold as a single card too. The provider had sat in the queue for two days as an address nobody could establish: geodd.com answers with a self-signed certificate, which we never bypass, and the real address, geodd.io, came from the aggregator's own link. No new card, and no figure of any other provider changed. The definition of the published figure did not change. | unchanged | |
| 1.32.5 | patch | Hetzner is read: one rate, the GEX131 dedicated GPU server at $2.2419 an hour, the NVIDIA RTX PRO 6000 Blackwell Max-Q's fifth source. The hourly figure under each server on its GPU matrix is drawn by a script, and the script states in the markup what it asks for - product-key="ROBOT_1747+ROBOT_1266", price-type="hourly" - so the module asks Hetzner's own keyless price endpoint for the same key. That key is one product to the endpoint, which answers it unchanged: the server together with the IPv4 address it is sold with. Nothing is added up here, because a sum of two products would be our arithmetic and the page prints none. The dollar figure is Hetzner's own, published beside the euro one, so this is a USD listing and not a conversion of ours. Scope is the page's own word, "starting from", which the product carries because it comes in three types. The one-off setup fee of $699 is not folded into the hourly rate: it is stated in the provider's note under the table, and the commitment column stays empty, because the server is billed by the hour. The second GPU server on the matrix, the GEX45, is not published: its RTX PRO 4000 Blackwell SFF Edition is a card this site has no id for, and one source is not the two a new card needs. No other figure changed. The definition of the published figure did not change. | unchanged | |
| 1.32.4 | patch | GPU.ai is read: 34 rates over 24 cards at 10:11 UTC, the widest single source this site has added. Its page serves no figure at all - the table is drawn by a script - so it is rendered, and every rate carries the scope its own footnote states, "lowest live on-demand rate per GPU, refreshed continuously": a floor for the card, not the price of one configuration, as at GMI Cloud, Sesterce, Neysa and Hyperbolic. It is read as GPU.ai's own rate card and not as a broker's quote of somebody else's, the test Hyperbolic passed on September 27: no other operator's price is printed anywhere on the page, no other operator is even named, and you rent from and pay GPU.ai, which bills per second and says the catalogue rate is the rate you are billed. Each card is offered in up to two tiers, and the tier word the row prints is the market: a community rate and a plain one are two quotes, unless they carry the same figure, which is the same figure twice and is dropped. The catalogue reprices within the hour, so each hourly read publishes what the table says at that minute, as a marketplace floor does. No new card and no figure of any other provider changed. The definition of the published figure did not change. | unchanged | |
| 1.32.3 | patch | The NVIDIA GeForce RTX 5060 Ti (16 GB) joins the catalogue with two sources, the same run and the same shape as the RTX 4070 Ti SUPER: Salad's rate card at $0.15 an hour at High priority ($0.123 / $0.097 / $0.07 down its preemptible tiers) and the Clore.ai floor at $0.1531 over 10 verified single-GPU listings at 04:20 UTC, the cheapest five $0.1304-$0.1641. The memory window divides editions rather than names here: the marketplace sells an 8 GB and a 16 GB card under this one name (16 machines and 34), only the 16 GB card has an id on this site, and the 8 GB machines stay out of the floor and are reported unmapped. Salad's table prints 16 GB beside its rate, which settles which card the figure is. The definition of the published figure did not change. | unchanged | |
| 1.32.2 | patch | The NVIDIA GeForce RTX 4070 Ti SUPER (16 GB) joins the catalogue with two sources, the shape of September 30: one published rate card and one marketplace floor. Salad prices it at $0.21 an hour at High priority and $0.17 / $0.13 / $0.09 down its three preemptible tiers; the Clore.ai floor is $0.2231 over 5 verified single-GPU listings at 04:20 UTC, the cheapest five $0.1750-$0.4375, every machine under the name reporting the 16 GB that is the card's only size. The two figures are 1.06x apart. Both readers carry a memory window, because the marketplace spells four cards of this family as four names and the RTX 4070 SUPER still has no id here. The definition of the published figure did not change. | unchanged | |
| 1.32.1 | patch | Civo's two Blackwell rate tables are read. Its pricing page prices every other card from a 1-GPU size and the B200 and B300 from one size only, an 8-GPU node, so a reader of "1 x" rows alone dropped both; the smallest size each table prices is now the quote, per node with its GPU count where that size is a node. Five new quotes, each a figure the page prints beside the commitment term it belongs to: B200 at 12 months $35.92 a node ($4.49 a GPU) and at 36 months $30.32 ($3.79), with no on-demand rate, its cell reading "N/A"; B300 on demand $59.92 ($7.49), at 12 months $51.92 ($6.49) and at 36 months $43.92 ($5.49). They are the first committed Blackwell rates on this site from anyone but Google Cloud and, for the B300, its second one-year rate. The fifteen quotes read before are unchanged. The definition of the published figure did not change. | unchanged | |
| 1.32.0 | minor | A monthly average of the hourly index over a fixed monthly panel, the value the index offers for a period (specification 1.2, section 2b). The panel of a month is every provider and region present in at least 90% of the hourly points of the 14 days before the 24th of the previous month, decided by the database and never changed; the month is the mean of the panel median at every hourly scrape, published at 75% coverage, frozen at the first scrape of the next month. September 2026 predates the rule and is averaged over every hourly point (panel open); October is the first fixed panel: 17 pairs from 15 providers, decided by the rule as of September 24. New files: /price-index-monthly.csv and /price-index/panel/2026-10.csv. The weekly index is unchanged. The definition of the published figure did not change. | unchanged | |
| 1.31.0 | minor | The Cloud GPU Price Index is published at every hourly scrape as well as weekly: the same rule (the newest USD on-demand H100 SXM listed rate of each provider and region read within the 48 hours before the scrape, marketplace floors out, at least 8 providers), computed by the database when the scrape is stored and written once with the quotes it was computed from. A week's value is its last hourly point rounded to the cent, so the weekly series is unchanged (the week of September 21 closed at $3.7025 in both, published $3.70). New files: /price-index-hourly.csv and one receipt per week under /price-index/hourly/; 588 of 590 points since September 6 were backfilled from the stored reads, two withheld (the first evening, one provider). The specification is at version 1.1. No average of the hourly points is published. The definition of the published figure did not change. | unchanged | |
| 1.30.27 | patch | Coverage: the NVIDIA RTX PRO 4500 Blackwell has a name on a marketplace for the first time on this site. Four rate cards price the card - EmpirioLabs, Massed Compute, RunPod's API and Amazon's g7 instances - and neither exchange we read had a name for it: Vast.ai carries none at all, and Clore.ai's went unmapped until today. It is read at 32 GB, which is the one size NVIDIA ships this card in and the size both machines on the marketplace report, so the question that keeps the RTX PRO 4000 Blackwell unmapped here does not arise for it: that card is sold as a full-height card and as a small-form-factor card at the same 24 GB, and neither its name nor its memory says which one a machine holds. Nothing is published from the new name today. Of the two machines carrying it at 22:04 UTC on 2026-09-30 one was rented, and the other, priced at $0.1308 an hour, carries no Verified badge - the marketplace's own mark for a host rated above four with reviews, and the line this site draws between an ask it will stand a floor on and one it will not. The book is withheld and reads thin market until three verified machines stand free under the name. The definition of the published figure did not change. | unchanged | |
| 1.30.26 | patch | Coverage: both editions of the NVIDIA RTX PRO 6000 Blackwell are read as one card on Clore.ai, and its lower-power Max-Q edition as the separate card it is. The marketplace prints Workstation Edition and Server Edition as two names over one book of 96 GB machines, and reading them as one card is not a new judgement: Vast.ai's two spellings of the same pair have been read as one card here since September, as have RunPod's and Massed Compute's. What was missing was a name map that can hold two names for one card. Without it the two names would have been two entries on one card, which would have taken two floors over two books and offered this site two quotes for one provider and one card - a duplicate, not a second source. The Max-Q is a different card with its own id, and since all three names report 96 GB it is the name alone that separates them, exactly as on the other marketplace. Nothing is published from any of the three today: all six single-GPU machines under them were rented at 22:04 UTC on 2026-09-30, and the multi-GPU book held two free verified four-card machines, at $5.95 and $5.7750 an hour, where a floor needs three. Both books are therefore withheld and read thin market until a third machine comes free. The definition of the published figure did not change. | unchanged | |
| 1.30.25 | patch | Coverage: the second marketplace on this site now has a name for the NVIDIA RTX PRO 5000 Blackwell, a card that until today stood on two sources here - RunPod's API and the Vast.ai floor. Clore.ai spells it out in full, and it is read with a memory window of 40 to 60 GB rather than by name alone, for the reason the other marketplace already gives: the same name is sold there in a 48 GB edition, which is the card this site holds, and a 72 GB edition, which has no id here and must not be priced under the smaller card's name. Nothing is published from it today. One machine carried the name at 22:04 UTC on 2026-09-30, and it was rented, so the reading is recorded as a withheld quote and the card page says thin market rather than showing a figure - a floor here needs three free verified machines at once, and it is the median ask of the five cheapest, never one machine and never an average. That is the same footing six datacenter names on this marketplace went in on the day before: a mapped name that publishes nothing today becomes a figure the hour the book turns over, and an unmapped one never does. The definition of the published figure did not change. | unchanged | |
| 1.30.24 | patch | Coverage: SwissGPU is read, and it is the first source on this site priced in Swiss francs. Seven dedicated servers, one card each, billed by the minute and priced by the hour: the RTX 4080 at 0.20 CHF an hour, the RTX 5080 and the RTX 3090 at 0.25, the RTX 4090 at 0.40, the RTX 5090 at 0.50, the RTX 6000 Ada at 1.00 and the RTX PRO 6000 at 1.50. As with every listing in a currency other than the dollar, the franc figure is what the page prints and what is shown beside the converted one, with the European Central Bank's rate for the day and its date: CHF 1 bought $1.1980 on 2026-09-30, which puts these seven between $0.24 and $1.80 a card-hour, each inside the normal band of its card. Electricity is not in those rates. The page repeats "+0.30 CHF/kWh" under every price and says in words that electricity is billed separately, so the figure is recorded as printed and the provider's note on the site says what it leaves out - the same treatment Thunder Compute's add-on cores get. That kilowatt-hour rate, and the half-franc fee for reinstalling an operating system, are never read as GPU prices. One detail of the page turned into a check worth keeping: it prints its price list twice, once narrow and once wide, so every card is read from both copies and a card priced differently in the two stops the read instead of one figure being chosen. The wide copy also names the architecture and the memory, which is what separates the RTX 6000 Ada at 48 GB from the RTX PRO 6000 at 96 GB - two cards one word apart on the same page - and a name whose memory does not match the card we hold is dropped rather than mapped. The definition of the published figure did not change. | unchanged | |
| 1.30.23 | patch | Coverage: SynpixCloud is read. Its catalogue prints no figure in the HTML it serves - the table arrives in the browser - so the page is rendered, and the reading otherwise follows every other source: our own name, one read an hour, a public page. What makes this provider different is that its table carries a stock column beside the price. Each card is sold as a one, two, four, six or eight GPU machine, and each of those rows says either Available or Out of Stock, with the Rent button disabled where it is out of stock. Only an available row is published, because a rate nobody can order is not a price a reader can act on - the same reading Vultr's eight-GPU H100 got on 2026-09-29, where the endpoint said the plan could not be deployed on demand. On the first read that keeps two well-known cards off the site entirely: every H100 row and every RTX 4090 row was out of stock, so this provider quotes neither, although the page prints $2.79 and $0.48 for them. What is published is one quote per card, the smallest machine of it that was available: the A100 80GB at $2.26 an hour, the A100 40GB at $0.79, the RTX 5090 at $0.79, the RTX A6000 at $0.69, the A40 at $0.63, the V100 32GB at $0.39, the V100 16GB and the RTX 3090 at $0.32, and the RTX 3080 as a two-card machine at $0.47, because its single-card row was the one out of stock. The larger machines are not published beside them: they work out under one percent cheaper per GPU, so they are the same product at the same price rather than a second offer, and a quote that hopped between five rows as stock turned over would read as a price move where nothing moved. A row whose memory is not its card's memory times its GPU count is dropped rather than mapped: this provider sells a 48 GB "RTX 4090PLUS" next to the 24 GB RTX 4090, and a familiar name carrying a different card is not a cheap one. That modified card stays unread for the same reason, as does the A800 40GB, an NVIDIA SKU of its own that no other source here prices. The definition of the published figure did not change. | unchanged | |
| 1.30.22 | patch | Coverage: Dataoorts is read. Its pricing page carries four separate rate cards, and names in its own script where three of them come from: a JSON endpoint behind the Nova NVLink VMs, two CSV files behind the Atlas and Yotta pods; the fourth, Orion, is a list written into the page. Those addresses are read out of the page rather than fixed in our code, so a file that moves is followed and a file that disappears stops that book instead of quietly shrinking it. Each series is published as a partition of its own, named beside the price the way RunPod's Secure and Community Cloud are, because they are different products rather than four ways to take one machine - a single-GPU VM with NVLink is not an eight-GPU bare-metal node. Collapsing them to one figure per card, which is what this module did first, published a spot price dearer than the on-demand price of the same card: Nova's H200 spot is $2.32 and Orion's eight-GPU node works out at $2.29 a card. Within a series the cheapest way to take the card stands, with the row's own name recorded. The Orion series is where that matters: its rows are not multiples of one another, and its eight-GPU bare-metal nodes are the cheapest H200 and RTX PRO 6000 on the page, $2.29 and $1.23 a card. What actually reached the site on the first read is six quotes from three of the four books - the B200 at $3.99, the RTX 4090 at $0.58, the RTX 5090 at $0.79, the RTX PRO 6000 at $2.19 and at $1.23 a card on an eight-GPU node, and the H200 at $2.29 a card the same way. The fourth book did not answer: the Nova endpoint is itself a proxy, and within ninety minutes of being read it went from serving nine cards to answering 503 and then 404, which is what the page shows as "Getting Real Time Cost..." where those figures belong. Its nine cards are therefore not published, the GB300 among them, so that card still stands on the two sources it had; if the endpoint comes back, its rates arrive with the next hourly read. That a book which does not answer is skipped rather than failing the whole provider is deliberate, and this is the case it was written for. Not published either: the X-Series, which sends a reader to the console instead of printing a rate; the marketplace, which is bidding behind a login; and the confidential-computing RTX PRO 6000, which would be a second figure of the same kind for a card already priced there. The definition of the published figure did not change. | unchanged | |
| 1.30.21 | patch | Coverage: Inferless is read, a serverless inference provider whose pricing page carries one plain table - hardware, type, a price per second and a price per hour, the GPU count, the vCPUs, the card memory and the host memory. Three quotes: the A100 80GB at $5.36 an hour, the A10 at $1.22 and the T4 at $0.66, each billed by the second and scaling to zero, so each is recorded as serverless. Every card on that page is priced twice, and only the Dedicated row of each is published: the Shared row beside it is half the same card - 40 GB against 80, 12 GB against 24, 8 GB against 16, with the vCPU count halved too - and a slice of a card is not a card, the reading Vultr's fractional vGPU rows got on 2026-09-29. The A100 Shared row is the one that could have passed for a card of its own, since a 40 GB A100 exists; what says it is a partition is the 80 GB printed on the same hardware name below it. Because one card is priced twice on the page, each quote carries the row's own name, so a reader following the link knows which of the two figures is ours. The two price columns cross-check each other - $0.001491 a second is $5.3676 an hour, printed as $5.36 - and only the hourly one is read. The A100 80GB is the dearest on the site at that figure, above Replicate's $5.04, which is what a serverless endpoint costs rather than a misreading. The definition of the published figure did not change. | unchanged | |
| 1.30.20 | patch | Coverage: the NVIDIA GeForce RTX 4070 (12 GB) joins the catalogue beside the RTX 5070 and on the same pair of sources - Salad's rate card at $0.15 an hour at High priority, and the Clore.ai floor $0.2625 over 6 verified single-GPU listings, the cheapest five $0.1527 to $0.2625 and every one of the six reporting 12 GB. AWS reads this name and drops it too, but its gr6 sizes are an L4 with more host memory rather than this card, so they are still not read. Both modules carry a memory window: the RTX 4070 Ti, on the site since 2026-09-29, and the RTX 4070 Ti SUPER are separate names with separate books on both sources, and a bare name would let one publish as the other. The definition of the published figure did not change. | unchanged | |
| 1.30.19 | patch | Coverage: the NVIDIA GeForce RTX 5070 (12 GB) joins the catalogue, the first of two cards today whose second source is a published rate card rather than a second marketplace. Salad prices it at $0.165 an hour at High priority, and the Clore.ai floor is $0.1969 over 17 verified single-GPU listings, the cheapest five $0.1750 to $0.2253 and every one of the 17 reporting 12 GB. The two figures are 1.19x apart, the closest a young consumer card has come on this site; the three cards that joined on 2026-09-29 stood on two marketplace floors each and one of those pairs was 2.2x apart. What changed is that Salad became readable. Both modules carry a memory window even though every machine in the book reported 12 GB: the RTX 5070 Ti is a different card with its own name and its own book on both sources, and the laptop editions carry the name as well. The definition of the published figure did not change. | unchanged | |
| 1.30.18 | patch | Coverage: Salad (SaladCloud) is read, the 64th provider with a live quote on the site, and the refusal that kept it off since 2026-09-13 is closed. That refusal said the price table ships empty and a script fills it from an API needing an organization key. Half of that is no longer true: the table does ship empty, but what fills it is a public static file the page loads with no key and no sign-in, carrying the whole rate card - each card's name, its memory, and a rate for each of the four priority tiers in the order the table's own head prints them. The figures a reader sees on the pricing page are therefore the figures recorded, which is why each row links to that page rather than to the file. Four quotes per card, one per tier, with the tier named in the region column, and the two price types are Salad's own words about them: High priority is "Never preempted by other workloads" and is read as on-demand, while Medium, Low and Lowest "may be preempted" and are read as spot. The market is community, the name Salad gives this capacity - idle consumer hardware, the shape the Clore.ai book has. Forty quotes on the first read across ten cards, every one matching the memory of our own id, and two cards that had only two sources gain a third: the RTX 3090 Ti at $0.19 an hour (beside RunPod $0.27 and the Vast.ai floor $0.4007) and the RTX 4070 Ti at $0.20 (between RunPod $0.19 and the Clore.ai floor $0.35). The RTX 5070 Ti gains a third source too, and its first that is not a marketplace floor. Not read: CPU-only container groups, the reserved Dedicated servers and volume deals that carry no published rate, and the classes the page sends to its portal "from $0.015 per GPU-hour", which is a floor for an unnamed set rather than any card's rate. The definition of the published figure did not change. | unchanged | |
| 1.30.17 | patch | Coverage: six names Clore.ai prints for cards this site already carries are now read rather than dropped - the A100 SXM4 40GB, the V100 SXM2 in both its sizes, the T4, the RTX A4000 and the RTX A6000. Five of the six are the mapping Vast.ai's module already makes, so the site's two marketplaces cannot read one name two ways; the sixth, the T4, the other marketplace has no name for, and NVIDIA shipped one T4 at one size. Each is a name plus a memory window, never a name alone. No figure is published by this today: none of the six books could carry a floor at the time of reading - nought to two free verified machines each, and every one of the 223 V100 SXM2 32 GB machines, the largest datacenter book on that marketplace, was rented or unverified - so each is recorded as withheld, and a card page shows nothing where a quote has never had a price. What the mapping buys is the next hour: a marketplace's book turns over, and a name nobody maps never becomes a figure. The definition of the published figure did not change. | unchanged | |
| 1.30.16 | patch | Coverage: Clore.ai's multi-GPU machines are now read as a second quote per card, the partition Vast.ai has always had - the machine's own price with its card count, region "Multi-GPU machines", never mixed into the single-GPU floor. Twelve node-priced quotes on the first run, and one card gains a source it needed: the RTX A2000, whose single-GPU book on this marketplace was one verified listing while ten whole machines of 6 to 12 cards stood beside it (Server #106239, $0.6965 an hour for 12 cards, $0.0580 a card). What had kept these machines unread was that a Clore host can let some of a machine's cards and keep the rest listed, which would make a node price something nobody can buy; the feed states it rather than leaving it to be inferred - false on a machine only ever rented whole, else a count of the cards still free - and a machine is priced only when every card in it is free. A machine holding two kinds of card is not any one card's node price and is read into neither book. The same two guards decide each book: a floor needs three machines, and a median over three times the cheapest ask is withheld as a thin market. The definition of the published figure did not change. | unchanged | |
| 1.30.15 | patch | Coverage: the NVIDIA GeForce RTX 3060 (12 GB) joins the catalogue, the cheapest card on the site and the second today whose two sources are both marketplaces - the Clore.ai floor $0.0678 over 6 verified single-GPU listings and the Vast.ai floor $0.0615 over 24, each the median of that marketplace's own five cheapest verified asks. No rate card on this site prices the card and RunPod's API carries no such name. Both books reported 12 GB throughout on the day, and both modules carry a memory window regardless, because the same name also covers an 8 GB card this site has no id for: a bare name would publish the smaller card as the larger one. The RTX 3060 Ti is a separate name and a separate book on both marketplaces. The definition of the published figure did not change. | unchanged | |
| 1.30.14 | patch | Coverage: the NVIDIA GeForce RTX 5070 Ti (16 GB) joins the catalogue with two sources, both of them marketplaces - the Clore.ai floor $0.2472 over 57 verified single-GPU listings and the Vast.ai floor $0.5363 over 5 verified offers, read within two minutes of each other. No other source on this site prices the card: RunPod's API carries no such name and it appears on no rate card we read. Each floor is the median of that marketplace's own five cheapest verified asks, so the 2.2x between them is two books of different hosts rather than a figure to average; the card page shows both and the reader can look either up. Laptop editions of the name are separate books on both marketplaces and never enter this one. The definition of the published figure did not change. | unchanged | |
| 1.30.13 | patch | Coverage: Taiga Cloud, the AI cloud of Northern Data Group, joins with two cards from its own pricing page - the H100 SXM at $2.60 an hour per GPU and the H200 SXM at $3.00. Both figures are printed as "from", and the page says why under each table ("Contact our team to unlock exclusive high-volume pricing"), so both are tagged as advertised floors rather than the price of a stated configuration, the tag Hyperstack's and GMI Cloud's "starting from" rates carry. The page prices each card twice, once as bare metal and once on demand, and these are two ways to take the same card by the hour rather than two markets, so the cheaper of the two is published and the row names which table it came from: the H100 is $2.60 either way, the H200 $3.00 bare metal against $3.20 on demand. The column heading is dollars per GPU-hour, so the figure is one card and not the eight-card machine the row describes. Taiga's vCPU table is not read: it prices processor cores and memory, not a GPU. Neither card gains a source it needed - the H100 SXM had 31 and the H200 40. The definition of the published figure did not change. | unchanged | |
| 1.30.12 | patch | Coverage: Vultr joins, read through its open plans endpoint api.vultr.com/v2/plans-metal, which answers a plain keyless request - Vultr's own website answers 403 to us on every path, the pricing page and the API reference alike, so the endpoint is where a reader checks these figures. The endpoint prices each bare-metal plan twice, on demand and preemptible, and says of each price whether the plan can be deployed that way; a rate is published only where it says it can. Today that is one on-demand plan, the four-card A100 80GB node at $9.589 an hour, $2.397 a card, and six preemptible ones, recorded as spot: the 8x H100 at $18.40 ($2.30 a card), 8x A100 SXM at $11.92 ($1.49), 8x L40S at $11.92 ($1.49), 8x MI300X at $14.80 ($1.85), 8x MI325X at $16.00 ($2.00) and 8x B200 at $25.60 ($3.20). Eight plans state an hourly cost while saying they cannot be deployed on demand - the 8x H100 at $23.92 among them - and those figures are not published as on-demand rates, because that is a label the source itself denies. A preemptible price equal to a plan's own on-demand price is the same figure twice rather than a second market and is left out, which is how the GH200, the four-card A100 and the 8x MI355X print theirs. Vultr's fractional cloud GPU plans, which sell slices such as 12GB of a 48GB A40, are not read: this site prices whole cards. No card gains or loses a source it needed - every card here already had three or more. The definition of the published figure did not change. | unchanged | |
| 1.30.11 | patch | Labelling: Together AI's reserved rates are now tagged "from" (advertised_from), because its GPU clusters page prints each one as "Reserved: Starting at" a figure per GPU-hour, above "Scale: 256 to 1,000+ GPUs" - a floor, not the price of a stated configuration. Until today they were tagged as instance prices; our price audit of 2026-09-29 found it, as it found the same with Hyperstack's "Starting from" reserved column on 2026-09-20. No figure moves: the reserved rates stay $3.19 for the H100 SXM, $6.79 for the B200 and $3.99 for the H200. Together's on-demand rates carry no such wording and keep their tag. Reserved rates do not enter the Price Index, so it is untouched. The definition of the published figure did not change. | unchanged | |
| 1.30.10 | patch | Coverage: the NVIDIA GeForce RTX 4070 Ti (12GB) joins the catalogue, a name RunPod's API has been reading and dropping since the card catalogue began. It has two sources today, each checkable where its row links: RunPod's API at $0.19 an hour in its Community Cloud, and the Clore.ai floor at $0.1991 over ten verified single-GPU listings, every one of them a 12GB machine, which is the memory RunPod reports for the card too. What changed is that Clore.ai became readable at all this morning; Vast.ai is not a third source, its book for the name holding one verified offer on 2026-09-27 and none on 2026-09-28. RunPod's Secure Cloud cell for the card is exactly $0.50, the placeholder its pricing page does not print, and it is skipped rather than published. A card with two sources shows a thin range, and says so. The definition of the published figure did not change. | unchanged | |
| 1.30.9 | patch | Coverage: Clore.ai joins as the site's second marketplace, and five cards that had only two sources now have three - the RTX 3070, RTX 3080 Ti, RTX 4080, RTX 4080 SUPER and RTX 5080. It is read on exactly the footing Vast.ai is: each figure is a trimmed floor, the median of the five cheapest verified, rentable, single-GPU listings, and it is withheld rather than published when the book holds fewer than three listings or the median runs more than three times the cheapest ask - which is why the RTX 3090 Ti and RTX A2000 read as a thin market on the first run, at one verified listing each. Verified is the marketplace's own badge, "rating above 4 with reviews", not a rule of ours. Clore prices a machine per day and its own table divides by 24 and adds the renter fee before showing a price, so the published figure is the listed price plus that fee over 24 hours; the fee is read from the marketplace on every run, 5% on 2026-09-29, and a run that cannot read it publishes nothing rather than guess. First figures, 04:40 UTC: RTX 4090 $0.3369, RTX 5090 $0.4856, RTX 3090 $0.1662, RTX 5080 $0.2927, RTX 4080 SUPER $0.3719, RTX 4080 $0.3063, RTX 3080 Ti $0.2188, RTX 3080 $0.1553, RTX 3070 $0.1194. Its bid-driven spot market is not read, and neither are multi-GPU machines, so no share of a larger machine is published as a machine. The definition of the published figure did not change. | unchanged | |
| 1.30.8 | patch | Coverage: the 32GB Tesla V100 is now a card of its own, v100-32, and three sources that had been reading its name and dropping it now publish it. The V100 this site has tracked since 2026-09-08 is the 16GB card, so a 32GB listing had nowhere to go; each of the three names the memory on its own page, which is what lets the two be told apart. CloudRift lists a V100 SXM3 at $0.28 an hour on demand and $0.25 for a month, beside the 16GB V100 SXM2 at $0.25 that stays the old card. Paperspace lists an eight-card V100-32G node at $9.20 an hour, $1.15 a card - the rate that was published as the 16GB card from 2026-09-08 until it was withdrawn on 2026-09-28. immers.cloud lists a four-card machine at 382.57 roubles an hour, 95.64 a card, $1.13 at the Bank of Russia rate of the day. No form factor is recorded for the new card: one source names SXM3 and the other two name none. Paperspace's other V100 row, printed as 'V100 / V100-32G', names either card and is published as neither. No existing figure changed hands: every quote that was on the site stays on the card it was on. The definition of the published figure did not change. | unchanged | |
| 1.30.7 | patch | Coverage: Cyfuture AI is no longer read, one day after it was added in 1.30.5, and its five rupee rates are withdrawn. The page that carries them (cyfuture.ai/pricing) answers 403 Forbidden to the machine that collects prices every hour, so not one of those figures was ever read on the site; it still answers a browser, and we do not work around a server that refuses us. Five cards lose this source: H100 SXM, A100 80GB, MI300X, L40S and RTX PRO 6000, each of which keeps every other source it had. No published figure changed, because none of Cyfuture's ever reached a page. The definition of the published figure did not change. | unchanged | |
| 1.30.6 | patch | Coverage: Selectel, a Russian cloud, is now read from its GPU page (selectel.ru/services/gpu/), which prices cloud servers by the hour in roubles on tariff cards. Only the cloud-server cards are read: the same page also prices managed Kubernetes node groups, at their own lower rate, and dedicated servers by the month, and neither is the product published here. Every figure on the page is printed as a 'from' rate and is labelled as such. One rate per card is kept, the machine with the lowest price per GPU, with its GPU count, converted at the Bank of Russia's official rate of the day. Three cards on 2026-09-28 at 84.41 roubles to the dollar: an H200 at 5,397.3 roubles an hour for eight cards ($7.99 a card), an RTX PRO 6000 at 579.94 for two ($3.44 a card) and an L4 at 69.88 ($0.83). As a rouble listing it does not enter the Price Index. The definition of the published figure did not change. | unchanged | |
| 1.30.5 | patch | Coverage: Cyfuture AI, an Indian cloud, is now read from its pricing page (cyfuture.ai/pricing), which lists whole GPU instances by the hour in rupees, one table per card. Only the on-demand column is read - the 1, 6 and 12-month reserved columns are a commitment - and only from a cell printed in rupees, because the page's own currency switch reports a failure. One rate per card is kept, the instance with the lowest price per GPU, with its GPU count, converted at the European Central Bank's reference rate for the rupee. Five cards on 2026-09-28 at 95.98 rupees to the dollar: an H100 SXM at 2,552 rupees an hour for eight cards ($3.32 a card), an A100 80GB at 1,536 for eight ($2.00 a card), an MI300X at 2,125 for eight ($2.77 a card), an L40S at 485 for four ($1.26 a card) and an RTX PRO 6000 at 256.5 for one ($2.67). The page also prices a B300 with no on-demand rate, a 32GB V100, an MI325X whose stated memory is not that card's, and an Intel Gaudi 2; none of those is published. As a rupee listing it does not enter the Price Index. The definition of the published figure did not change. | unchanged | |
| 1.30.4 | patch | Coverage: Cloud.ru Evolution, the largest Russian public cloud, is now read from the GPU tariff annex of its contract (cloud.ru/documents/tariffs/evolution/evolution-compute-gpu, version 260909), which prices 41 GPU virtual machines by the hour in roubles. The annex prints each rate with VAT, without VAT and the VAT itself; the figure published is the one with VAT, what a buyer pays. One rate per card is kept, the machine with the lowest price per GPU, with its GPU count, converted at the Bank of Russia's official rate of the day. Six cards on 2026-09-28 at 84.41 roubles to the dollar: an A100 40GB at 256.2 roubles an hour ($3.04), an A100 80GB at 732 for two cards ($4.34 a card), an H100 PCIe at 549 ($6.50), an H100 NVL at 646.6 ($7.66), an H100 (form factor not stated) at 1,708 for two ($10.12 a card) and a V100 at 244 ($2.89). The annex also prices an A100 PCIe without saying how much memory it has, on eleven cheaper rows; those are not published, because the card cannot be told from the page. As a rouble listing it does not enter the Price Index. The definition of the published figure did not change. | unchanged | |
| 1.30.3 | patch | Coverage: immers.cloud, a Russian GPU cloud, is now read from its GPU page (immers.cloud/gpu/), which lists ready-made virtual machines per card by the hour in roubles. One rate per card is published, the machine with the lowest price per GPU, with its GPU count, converted at the Bank of Russia's official rate of the day; the company states it pays no VAT. Ten cards on 2026-09-28 at 84.41 roubles to the dollar, among them an H100 (form factor not stated) at 1,312.65 roubles an hour for four cards ($3.89 a card), an H200 at 839.97 for two ($4.98 a card) and an A100 80GB at 792.65 for four ($2.35 a card). As a rouble listing it does not enter the Price Index. The definition of the published figure did not change. | unchanged | |
| 1.30.2 | patch | Coverage: Intelion Cloud, a Moscow GPU cloud, is now read from its server configurator (intelion.cloud/gpu_server_configurator/), the first source priced in roubles. The page prices the card alone, VAT included, with CPU, memory and disk added on top, so each rate is tagged as GPU only and shown as printed, converted at the Bank of Russia's official rate of the day. Four cards on 2026-09-28 at 84.41 roubles to the dollar: A10 27.38 roubles an hour ($0.32), RTX A5000 37.20 ($0.44), RTX 3090 50 ($0.59), A100 80GB 204.17 ($2.42). Its H100 SXM and H200 are not read: they cannot be ordered from the configurator and their own pages print different figures and say the card is not available. As a rouble listing it does not enter the Price Index. The definition of the published figure did not change. | unchanged | |
| 1.30.1 | patch | Coverage: ionstream is no longer read. Its page prints one "from" rate per GPU-hour for the B200 ($2.69) and the B300 ($2.93) with no term, and we published both as on-demand - the cheapest of each card on the site. The company has since told us in writing that these rates assume a large part of the server paid up front and that the B200 is no longer available; the page states neither, so the rates are withdrawn rather than relabelled, and the figures in the letter are not published. The cheapest listed on-demand rate is now $3.69 for the B200 and $4.89 for the B300. The definition of the published figure did not change. | unchanged | |
| 1.30.0 | minor | A price listed in any currency can now be published (owner's decision): until today only USD and EUR. The listed figure is shown as listed, next to its USD figure at the day's official reference rate and that rate's date - the ECB rate (for a currency other than the euro, the cross of the two rates the ECB publishes against the euro), or the Bank of Russia's official rate for RUB and VND, which the ECB does not publish. Price changes are still detected on the listed figure, never on the rate, and only USD listings enter the Price Index. No quote changed today: sources priced only in roubles, yen, rupees or dong become readable from the next parser run. The definition of the published figure did not change. | unchanged | |
| 1.29.6 | patch | Coverage: the row Hyperstack prints as "NVIDIA H100 NVLink" is now read, the last of the six H100 and A100 rows on that page that was still being dropped. Its VRAM column says 80 GB, so it is the PCIe card on an NVLink bridge rather than the 94 GB H100 NVL, which is how the "NVIDIA A100 NVLink" row has been read since the page was first scraped. No published figure moves: Hyperstack prices the plain "NVIDIA H100" row lower in both sections ($2.50 an hour against $2.60 on-demand, $1.75 against $1.82 reserved), and the cheapest of the rows for a card is the one published, with the name it was read under shown beside it. What changes is that the rate is now read at all, so it becomes the quote if Hyperstack ever stops listing the plain row. The definition of the published figure did not change. | unchanged | |
| 1.29.5 | patch | Coverage: Together AI's HGX B300 card is now read, at $9.99 per GPU-hour on-demand. The card had carried "Contact us for pricing" and no figure; it now prints an hourly rate, and the page prints no "Starting at" reserved line beside it, so this is the first Together card published with an on-demand rate and no reserved one. The 270 GB HGX name is the eight-way SXM board, read as the same B300 the rest of the site tracks. The GB200 NVL72 card on the same page still says "Contact us for pricing" and is not read. The definition of the published figure did not change. | unchanged | |
| 1.29.4 | patch | Coverage: the row CloudRift lists as "H100" is now read, at $1.70 an hour on a one-month reservation. It had been dropped because the model column names no form factor; the row's own VRAM column says 80 GB, so it is published against the form-factor-neutral H100 card, which says exactly that and no more - the same footing Gcore's 1x H100 and Sesterce's "H100 - SXM5 / PCIe" are read on. Its On-Demand cell reads "Upon request" and publishes no figure, so the one-month rate is the only quote this row carries. CloudRift's "V100 SXM3" stays unread: 32 GB is another card, not another name for one we track. The definition of the published figure did not change. | unchanged | |
| 1.29.3 | patch | Coverage: ionstream joins as the 57th source read, with two rates from the line-up on its solutions page - B200 $2.69 and B300 $2.93 per GPU-hour. Both lines say "from", so both are tagged as advertised "from" prices rather than the price of a configuration, the same footing as GMI Cloud, Sesterce and Neysa. They are read as on-demand because the page names no term, no minimum and no commitment: its own words are "Pick the GPU that matches your workload and scale on demand". Each figure is already per GPU on the page, so nothing is divided; the eight-card node it describes carries no total price. The Vera Rubin card on the same page is "coming soon" with no figure and is not read. Both rates are the lowest this site reads for their card, which is what an advertised "from" is, and each is tagged so a reader can see it. The definition of the published figure did not change. | unchanged | |
| 1.29.2 | patch | Coverage: Hyperbolic joins as the 56th source read, with three rates from the card on its marketplace page - H100 SXM $3.19, H200 $3.99 and B200 $5.99 per GPU-hour. The column they sit under says "Starting From", so all three are tagged as advertised "from" prices rather than the price of a configuration, the same footing as GMI Cloud, Sesterce and Neysa: the page itself says the figures are refreshed weekly from the best available rates from suppliers on its platform. They are read as Hyperbolic's own rate card, not as a broker quoting another operator: no other operator's price appears anywhere on the page, and the comparison table that names AWS, Azure, CoreWeave, Fluidstack, Lambda Labs and RunPod prints multiples and no rates, so nothing in it can become a figure here. Each of the three sits inside the live range of its card. The definition of the published figure did not change. | unchanged | |
| 1.29.1 | patch | Coverage: Gcore joins as the 55th source read, with nine rates from its AI pricing page, which carries three separate GPU products. Its bare-metal clusters are priced per machine and published that way - EUR 18.16 an hour for 8x H100 SXM, EUR 10.16 for 8x A100 SXM, EUR 8.72 for 8x L40S PCIe, EUR 21.84 for 8x H200 SXM and EUR 48.00 for 8x B300 SXM - and a column is published only when the per-GPU figure derived from the node rate equals the one Gcore prints beside it. Everywhere Inference, an endpoint rather than a machine by the page's own words, adds EUR 1.08 for 1x L40S, EUR 1.26 for 1x A100 and EUR 1.53 for 1x H100 as serverless rates; Virtual GPU adds EUR 2.36 for 1x H100. The rates are listed in euro and published in euro beside the converted figure with the ECB rate and its date, so, like every EUR listing, they stay out of the Price Index. Which of the three products a rate comes from is shown on the row, because the same H100 costs a different amount in each. The 72x GB300 NVL72 column prints no rate at all and is not read. Only the one-GPU row of each per-configuration table is read; the larger rows are that same rate multiplied. The definition of the published figure did not change. | unchanged | |
| 1.29.0 | minor | A new market, shared: a GPU other tenants use too, first packet.ai's Dynamic plan ("shared infrastructure with the same peak performance and VRAM" at "roughly half the price"). Like community capacity it counts in no median rate card and no "cheapest" figure, and its risk tag is medium; it had been read as on-demand since 2026-09-26, when its RTX PRO 6000 at $0.66 became the cheapest card from 80 GB against $1.09 next. The Price Index is untouched - no shared quote is an H100 - and so is every other figure's definition; the change is which rates count as an on-demand rate card. | unchanged | |
| 1.28.29 | patch | Coverage: Deep Infra joins as the 54th source read, with five rates from the "Custom LLMs" table on its pricing page - A100 80GB $0.89, H100 $2.20, H200 $2.69, B200 $3.69 and B300 (270 GB) $4.89 an hour, each printed there as the provider's own "$X / GPU-hour". The rest of that page is per-token model pricing, which this site does not track, and the DGX clusters below the table are contact-sales; neither is read. The rates are recorded as on-demand instances: the page says "pay for uptime" with "minute granularity" and names no term, and nothing else on it is billed by the hour. "H100 80GB" states no form factor, so it is the neutral H100 and stays out of the H100 SXM index. All five sit at or just under the cheapest rate we read for their card, which is where an inference platform selling capacity by the minute belongs. The definition of the published figure did not change. | unchanged | |
| 1.28.28 | patch | Coverage: the NVIDIA RTX A2000 (6 GB) joins the catalogue as the smallest card on the site, with two sources - RunPod $0.12 an hour in its Community Cloud, read from RunPod's API as every RunPod row is and the cheapest rate it prints for anything, and the Vast.ai marketplace floor, $0.2037 an hour over 3 verified single-GPU offers on 2026-09-27, every one of them reporting 6 GB. RunPod's Secure Cloud cell for the card is exactly $0.50, the placeholder figure no RunPod page prints, and is skipped. The marketplace name covers two cards, so it is read below 8 GB: the cheapest ask under the bare name that morning was a 12 GB A2000 at $0.0521, a card this site has no id for. Three offers is the smallest book a floor is taken over here, and the same book was refused four hours earlier for a floor 3.6 times its cheapest ask, so this quote will read "thin market" on some hours - the guard working, not a gap. The definition of the published figure did not change. | unchanged | |
| 1.28.27 | patch | Coverage: AWS becomes a source for the NVIDIA RTX PRO 6000 Blackwell Server Edition (96 GB), at $3.3631 an hour for a g7e.2xlarge in us-east-1. G7e is the other half of the G7 generation mapped six hours earlier, and a different card: AWS's own G7e page says "G7e instances feature up to 8 NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs with 768 GB of total GPU memory (96 GB of memory per GPU)" and its size table opens with "g7e.2xlarge | 1 | 96", so the GPU count is read and not derived. Three of its sizes carry one card and the smallest is the cheapest per GPU, as with every other AWS family read here ($3.9982 for g7e.4xlarge, $5.2682 for g7e.8xlarge). The figure sits among the dearest of the card's quotes, between Modal $3.0312 and Oracle $4.50; the Max-Q edition is a card of its own here and AWS prices no such size. The definition of the published figure did not change. | unchanged | |
| 1.28.26 | patch | Coverage: AWS becomes a fourth source for the NVIDIA RTX PRO 4500 Blackwell (32 GB), at $2.52 an hour for a g7.2xlarge in us-east-1. The G7 generation had sat unmapped in AWS's pricing JSON since it appeared; AWS's own G7 page names the card - "Up to 8 NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs per instance with 32 GB memory per GPU" - and its size table gives one GPU for a g7.2xlarge, so the count is read and not derived. It is the dearest of the card's four quotes, beside EmpirioLabs $1.40, Massed Compute $0.92 and RunPod $0.72 in its Secure Cloud, which is where AWS sits for every card we read from it. The g7e generation is the RTX PRO 6000 Blackwell Server Edition and stays unmapped. The definition of the published figure did not change. | unchanged | |
| 1.28.25 | patch | Coverage: the NVIDIA A10G (24 GB) joins the catalogue with two sources - AWS at $1.006 an hour for a g5.xlarge in us-east-1, and Baseten at $1.2072 an hour, the $0.02012 a minute its pricing page prints for an "A10G 24 GiB VM". Both parsers had been reading the card and dropping it. It is AWS's own variant of the A10 and not the A10, which this site tracks separately and prices from $0.59 to $3.20 an hour; the two never share an id. The number of cards in a g5.xlarge is read off AWS's own G5 instance table, which prints one GPU of 24 GB for that size, and as with every AWS family only the smallest size is read, being one whole card and the cheapest per GPU. The g5g family stays out: it is the T4G, another card. The definition of the published figure did not change. | unchanged | |
| 1.28.24 | patch | Coverage: the NVIDIA GeForce RTX 3090 Ti (24 GB) joins the catalogue with two sources - RunPod $0.46 an hour in its Secure Cloud and $0.27 in its Community Cloud, read from RunPod's API as every RunPod row is, and the Vast.ai marketplace floor, $0.2578 an hour over 5 verified single-GPU offers on 2026-09-27, every one reporting 24 GB, with a four-GPU machine at $1.7119 an hour for the node as a second quote. Neither RunPod cell is the $0.50 placeholder, so the card arrives with two rates rather than one. It is a different card from the GeForce RTX 3090, and because both hold 24 GB the names keep them apart rather than the memory: Vast.ai sells it as "RTX 3090 Ti" and RunPod as "NVIDIA GeForce RTX 3090 Ti", each with its own book and its own rate. Both parsers had been reading the name and dropping it. The definition of the published figure did not change. | unchanged | |
| 1.28.23 | patch | Coverage: the NVIDIA GeForce RTX 4080 SUPER (16 GB) joins the catalogue with two sources - RunPod $0.28 an hour in its Community Cloud, read from RunPod's API as every RunPod row is, and the Vast.ai marketplace floor, $0.3511 an hour over 5 verified single-GPU offers on 2026-09-27, every one reporting 16 GB, with a two-GPU machine at $0.5356 an hour for the node as a second quote. RunPod's Secure Cloud cell for the card is exactly $0.50, the placeholder figure no RunPod page prints, and is skipped. It is a different card from the GeForce RTX 4080 published the day before, and because both hold 16 GB the names keep them apart rather than the memory: Vast.ai sells it as "RTX 4080S" and RunPod as "NVIDIA GeForce RTX 4080 SUPER", each with its own book and its own rate. Its Vast.ai book carried two verified offers on 2026-09-26, where a floor needs three, and five the following night. The definition of the published figure did not change. | unchanged | |
| 1.28.22 | patch | Coverage: the NVIDIA GeForce RTX 3080 Ti (12 GB) joins the catalogue with two sources - RunPod $0.18 an hour in its Community Cloud, read from RunPod's API as every RunPod row is, and the Vast.ai marketplace floor, $0.1889 an hour over the five cheapest of 9 verified single-GPU offers on 2026-09-27, every one reporting 12 GB, with a two-GPU machine at $0.3756 an hour for the node as a second quote. RunPod does not offer the card in its Secure Cloud at all, so there is one rate and not two. It is a card of its own beside the GeForce RTX 3080 published the same hour - 12 GB against 10 - and the two are never folded together: each has its own name at both sources and its own book on the marketplace. The definition of the published figure did not change. | unchanged | |
| 1.28.21 | patch | Coverage: the NVIDIA GeForce RTX 3080 (10 GB) joins the catalogue with two sources - RunPod $0.17 an hour in its Community Cloud, read from RunPod's API as every RunPod row is, and the Vast.ai marketplace floor, $0.1344 an hour over the five cheapest verified single-GPU offers on 2026-09-27, every one reporting 10 GB, with a three-GPU machine at $0.4230 an hour for the node as a second quote. RunPod does not offer the card in its Secure Cloud at all, so there is one rate and not two. Both parsers had been reading the name and dropping it. Vast.ai sells 10 GB and 20 GB machines under the one name - one of the 16 verified offers that hour reported 20 GB - so its book is read below 16 GB and the 20 GB card, which has no id here, is not priced under this one. The definition of the published figure did not change. | unchanged | |
| 1.28.20 | patch | Coverage: the NVIDIA GeForce RTX 3070 (8 GB) joins the catalogue with two sources - RunPod $0.13 an hour in its Community Cloud, read from RunPod's API as every RunPod row is, and the Vast.ai marketplace floor, $0.0978 an hour over 10 verified single-GPU offers on 2026-09-26, every one reporting 8 GB, with a two-GPU machine at $0.3489 an hour for the node as a second quote. RunPod does not offer the card in its Secure Cloud at all, so there is one rate and not two. Both parsers had been reading the name and dropping it. The RTX 3070 Ti is a card of its own and is published by neither source. The definition of the published figure did not change. | unchanged | |
| 1.28.19 | patch | Coverage: the NVIDIA GeForce RTX 4080 (16 GB) joins the catalogue with two sources - RunPod $0.27 an hour in its Community Cloud, read from RunPod's API as every RunPod row is, and the Vast.ai marketplace floor, $0.2810 an hour over 5 verified single-GPU offers on 2026-09-26, every one reporting 16 GB. Both parsers had been reading the name and dropping it since we started; what changed is the Vast.ai book, which carried no verified offer at 04:20 UTC and five at 16:12. RunPod's Secure Cloud cell for the card is exactly $0.50, the placeholder figure no RunPod page prints, and is skipped. The RTX 4080 SUPER is a different card, listed under its own name by both sources and published by neither: its Vast.ai book had two verified offers, and a floor needs three. The definition of the published figure did not change. | unchanged | |
| 1.28.18 | patch | Coverage: RunPod becomes the fourth source for the NVIDIA RTX PRO 6000 Blackwell Max-Q (96 GB), at $1.64 an hour in its Community Cloud, read from RunPod's API as every RunPod row is. NVIDIA names all three editions of the card in full, so the name RunPod carries - "RTX PRO 6000 Blackwell Max-Q Workstation Edition" - is this card and not the Server or Workstation edition we already read under another id. RunPod's Secure Cloud cell for it is exactly $0.50, the placeholder figure no RunPod page prints, and is skipped as every $0.50 cell is; the one rate left, $1.64, sits between the three the card's other sources published the same day - the Vast.ai community floor $1.34, CloudRift $1.40 for a month and EmpirioLabs $1.80 on demand. The definition of the published figure did not change. | unchanged | |
| 1.28.17 | patch | Coverage: Azure's RTX PRO 6000 Blackwell Server Edition is read - the two sizes Microsoft's size table (NC RTX PRO 6000 BSE v6 series) gives one whole GPU each, NC144ds_xl and NC144lds_xl, on demand, spot and at one- and three-year reservations, from the Retail Prices API (first read $5.50 an hour on demand, eastus2). The quarter- and half-card sizes stay out (a fraction of a GPU is not a GPU-hour, the A10 rule), as do the two-GPU NC288 sizes (the same price per GPU) and the sizes the table does not list; all are reported as unmapped. The card is not an H100, so the Price Index is untouched; the definition of the published figure did not change. | unchanged | |
| 1.28.16 | patch | Coverage: the NVIDIA RTX 5000 Ada Generation (32 GB) joins the catalogue with two sources - RunPod $0.83 an hour in its Secure Cloud and $0.49 in its Community Cloud, read from RunPod's API as every RunPod row is, and the Vast.ai marketplace floor, $0.4548 an hour over 3 verified single-GPU offers on 2026-09-26, all reporting 32 GB. Both parsers had been reading the name and dropping it. It is a card of its own, neither the RTX A5000 (24 GB, Ampere) nor the RTX PRO 5000 Blackwell (48 GB), and the three are priced apart. The definition of the published figure did not change. | unchanged | |
| 1.28.15 | patch | Coverage: Leafcloud joins the rate cards - an Amsterdam provider that re-uses its machines' heat in buildings, listing in euro. Four figures on 2026-09-26, per GPU per hour and VAT excluded: RTX 6000 Blackwell EUR 2.76, H100 EUR 3.15, A100 80GB EUR 1.61 and A30 EUR 0.60, each shown beside the ECB conversion and its date as every euro listing is, and each outside the Price Index as every euro listing is. Leafcloud prints its rate card by the month and its hourly rates in a comparison table beside other providers'; we publish its own hourly row only, and only after checking each figure against the same flavor's monthly price over 730 hours and its card count. Its H100 names no form factor, so it is the neutral H100 and does not enter the H100 SXM index. The definition of the published figure did not change. | unchanged | |
| 1.28.14 | patch | Coverage: packet.ai joins the rate cards, read from its own static pricing page. Six figures on 2026-09-26 - L40S $0.92, RTX 4090 $0.39, A100 80GB $1.43 and B200 $6.99 an hour on its Dedicated plan (a whole card, single-tenant), RTX PRO 6000 $0.66 and B200 $3.75 on its Dynamic plan (a shared card at the same peak performance and VRAM, by packet.ai's own description). The plan is the row's region, the way RunPod's Secure and Community Cloud are; both plans are billed hourly and neither is pre-emptible, so both are on-demand. The cards whose plans read "Launching soon" - RTX 5090, H100 SXM, H200 - and the Clusters plan, which is quoted rather than priced, are not published, and the comparison table of other providers' rates on the same page is not read at all. The definition of the published figure did not change. | unchanged | |
| 1.28.13 | patch | Coverage: the NVIDIA GeForce RTX 5080 (16 GB) joins the catalogue with two sources - RunPod $0.59 an hour in its Secure Cloud and $0.39 in its Community Cloud, read from RunPod's API as every RunPod row is, and the Vast.ai marketplace floor, $0.3356 an hour over 8 verified single-GPU offers on 2026-09-26, all reporting 16 GB. Both parsers had been reading the name and dropping it. RunPod's other GeForce cards - the RTX 4080, 4080 SUPER, 3080 and the rest - stay unread until a second source prices them. The definition of the published figure did not change. | unchanged | |
| 1.28.12 | patch | Coverage: the NVIDIA RTX PRO 5000 Blackwell (48 GB) joins the catalogue with two sources - RunPod $0.96 an hour in its Secure Cloud and $0.82 in its Community Cloud, read from RunPod's API as every RunPod row is, and the Vast.ai marketplace floor, $0.6434 an hour over 18 verified single-GPU offers on 2026-09-26. Both parsers had been reading the name and dropping it. Vast.ai sells a 72 GB edition under the same name, which is a different card with no id here, so its offers are kept out of this card's floor by a memory window. The definition of the published figure did not change. | unchanged | |
| 1.28.11 | patch | Coverage: Vast.ai is a third source for the NVIDIA RTX PRO 6000 Blackwell Max-Q (96 GB), whose floor was $1.3716 an hour over 9 verified single-GPU offers ($1.0030-$1.7370) on 2026-09-26, beside EmpirioLabs $1.80 and CloudRift $1.40 for a month. The marketplace calls it "RTX PRO 6000 Max-Q" and it is read as a book of its own: none of its offers reach the RTX PRO 6000 floor, which is taken over the Server and Workstation editions only. The definition of the published figure did not change. | unchanged | |
| 1.28.10 | patch | Coverage: the NVIDIA RTX 4000 SFF Ada (20 GB) joins the catalogue with two sources - Koyeb $0.50 an hour for its serverless instance, RunPod $0.44 in its Secure Cloud and $0.18 in its Community Cloud, read from RunPod's API as every RunPod row is. The SFF edition is a 70 W half-height card and is kept apart from the full-height RTX 4000 Ada listed by DigitalOcean, RunPod and Akamai: the two ids never mix, and neither card enters the other's figures. Both parsers had been reading this name and dropping it. The definition of the published figure did not change. | unchanged | |
| 1.28.9 | patch | Coverage: the NVIDIA RTX PRO 6000 Blackwell Max-Q (96 GB) joins the catalogue with two sources - EmpirioLabs $1.80 an hour on demand, CloudRift $1.40 an hour on a one-month commitment, its on-demand cell reading "Upon request". The Max-Q is the lower-power edition of the RTX PRO 6000 and is a card of its own here: its figures never enter the RTX PRO 6000's, and both parsers had been reading the name and dropping it. RunPod's Max-Q cells stay out, its secure price being the $0.50 placeholder the API returns in unrelated cells - below its own $1.64 community price. The definition of the published figure did not change. | unchanged | |
| 1.28.8 | patch | Coverage: the NVIDIA RTX PRO 4500 Blackwell (32 GB) joins the catalogue with three sources, a name three of our parsers already read and dropped - EmpirioLabs $1.40 an hour, Massed Compute $0.92, RunPod $0.72 in its Secure Cloud and $0.34 in its Community Cloud, the last read from RunPod's API as every RunPod row is. RunPod lists the card in two editions, plain and Server; they are the same card, as the RTX PRO 6000 pair already is, and the Server Edition's $0.50 community cell is the placeholder the API returns in unrelated cells, so it is not published. The definition of the published figure did not change. | unchanged | |
| 1.28.7 | patch | Coverage: Enverge (RTX Pro 6000 $1.95 and B300 $7.50 for a single-GPU machine) and EcoHash (RTX Pro 6000 $1.89 per GPU-hour) are read. Two candidates are listed as not listed, with the reason: BreezeHost (one card, out of stock, its on-demand rate only in the page title) and 1Legion (the page prints two different 12-month rates for the same card, so neither can be taken as its price). The definition of the published figure did not change. | unchanged | |
| 1.28.6 | patch | Coverage: six more providers from the survey of competitors' lists, each a public rate card. Lyceum (its on-demand VM table: H100 $2.79, H200 $4.29, B300 $7.99), Canopy Wave (H200 $4.00, H100 SXM $3.25 per GPU, from two-GPU machines), Atlas Cloud (H200 $3.50, H100 SXM $2.95), Nova Cloud (RTX 4090 $0.42, RTX 5090 $0.66, RTX PRO 6000 $1.56), Neysa (four cards marked "starts at", so shown as advertised floors: H100 SXM $4.39) and HexGrid (twelve cards, RTX A4000 $0.20 to B200 $7.50). Every H100 whose page names no form factor stays out of the H100 SXM figures. The definition of the published figure did not change. | unchanged | |
| 1.28.5 | patch | Coverage: Hugging Face Inference Endpoints are read - eight GPUs at the single-card size of a dedicated endpoint, on the cloud each row names (AWS today): T4 $0.50, L4 $0.80, L40S $1.80, A100 80GB $2.50, H100 $4.50, H200 $5.00, B200 $9.25, RTX PRO 6000 $2.75 an hour. An endpoint serves a model rather than giving a machine to log into, and the provider page says so. Its H100 names no form factor and stays out of the H100 SXM figures. The definition of the published figure did not change. | unchanged | |
| 1.28.4 | patch | Coverage: Daytona is read - eight GPUs of its sandbox pricing, on demand and preemptible (recorded as spot), each the GPU alone, since vCPU, memory and storage are billed on top: B300 and B200 $6.25, MI355X $5.99, H200 $4.54, H100 $3.95 on demand; H100 $2.27 preemptible. Its H100 names no form factor and stays out of the H100 SXM figures. The definition of the published figure did not change. | unchanged | |
| 1.28.3 | patch | Coverage: Civo is read - A100 40GB $1.09, A100 80GB $1.79, H100 SXM $2.99, H200 $3.49 and L40S $1.29 an hour on demand for one GPU with its CPU, memory and disk, and the same cards at 12- and 36-month commitments (H100 $2.69 and $2.49). Civo's 6- and 24-month terms have no market on this site and are not recorded. Its H100 is the SXM card its table names, so it enters the H100 SXM Price Index panel on the index's own rules. The definition of the published figure did not change. | unchanged | |
| 1.28.2 | patch | Coverage: EmpirioLabs is read - its GPU Cloud table of 21 cards, each billed per second at an hourly rate, among them H100 $2.60, A100 80GB $1.70, B300 $16.00. A row priced as an "8-GPU node" is kept as a node price with its count. Rows the page shows as "0 available" are published as listed; the site does not track stock. The GeForce RTX 3090 (24 GB) joins the catalogue with three sources: RunPod $0.22 community and $0.50 secure, Vast.ai's marketplace floor, EmpirioLabs $0.40. The definition of the published figure did not change. | unchanged | |
| 1.28.1 | patch | Coverage: Akamai Cloud (Linode) is read, from its documented keyless plan API - its pricing page refuses our client and is not read - and two cards join the catalogue, each with three sources. Quadro RTX 6000 (Turing, 24 GB): Seeweb EUR 0.29 ($0.33), Lambda $0.69, Linode $1.50. RTX 4000 Ada (20 GB): RunPod $0.20 community and $0.28 secure, Linode $0.52 for its smallest plan, DigitalOcean $0.76. The SFF edition of the RTX 4000 Ada is a different card and stays out. The definition of the published figure did not change. | unchanged | |
| 1.28.0 | minor | Two changes to what can be read, by the owner's decision. A card joins the catalogue when two providers publish its price verifiably, where three were needed before; its median, as for every card, needs two rate cards, so a card with two sources shows a thin but honest range. And a provider whose page draws its prices with JavaScript can now be read: a headless browser renders the page as a visitor's browser would, reads robots.txt first, identifies itself as our parser, and stops at any bot check or captcha - a page that asks is recorded as not listed, never worked around. No provider is read this way yet, and no published figure changed. | unchanged | |
| 1.27.0 | minor | A price read from a provider's API may be published although its public page does not print it, by the owner's decision - the row says "read from the provider's API" with the endpoint, as every RunPod, Vast.ai, AWS, Azure, Oracle, Cudo and Google Cloud row already does, so a reader checks it where it comes from. Until now a RunPod card was taken only if runpod.io/pricing printed it. First use: RunPod's RTX A4000 ($0.17 community, $0.25 secure) and H200 NVL ($3.79 secure). RunPod's API also returns exactly $0.50 in a dozen unrelated cells - the H200 NVL's community price beside $3.79, the RTX 2000 Ada's community price above its own secure one - so a $0.50 cell is published only where /pricing prints that price for that card and tier. The NVIDIA H200 NVL (141 GB, PCIe) joins the catalogue with three sources: Massed Compute $3.62, RunPod $3.79 and Vast.ai's floor $4.14. It is not the SXM H200 and never enters its figures. The definition of the published figure did not change. | unchanged | |
| 1.26.1 | patch | Azure's H200 spot quote is retired. Azure's price API lists a "Spot" meter for the H200 at exactly the on-demand price, in all 32 regions it prices the card - no discount, so no spot market to report; we had been publishing $10.60 per GPU-hour as spot beside the same $10.60 on demand. The rule is the one RunPod's identical spot rows were retired under on 2026-09-11: a spot price is published only where it differs from on-demand for the same machine in the same region. It also removes 19 of Azure's 20 GB200 spot meters, but not the one published (Korea South, $6.22, which does differ). No other published figure changed. | unchanged | |
| 1.26.0 | minor | Azure and Google Cloud rows carry the spread across the regions we read. Both are read in every region and the published row is the cheapest region per card and market, as before; the row now also says how many regions were read and what the dearest of them asks per GPU-hour, each region at its own cheapest offer - "cheapest of 38 regions read; dearest $7.61" for Azure's A100 80GB on demand ($3.67). Two new fields in prices.json and its CSV, regions_read and region_max_usd_per_gpu_hour, null for every other provider. The count is of regions read, not of the regions a provider has. Azure Government and Azure China regions are left out of both the spread and the pick; none had ever been the cheapest. No published figure changed. | unchanged | |
| 1.25.11 | patch | Coverage: AWS is read for the NVIDIA B300 - p6-b300.48xlarge at $142.416 an hour on demand in us-east-1, eight cards a node, $17.80 per GPU-hour. The instance sat unmapped in the same pricing JSON the other nine AWS rows come from; AWS prints one row for it, "8 Ultra" GPUs and 2,144 GB HBM3e, which is 268 GB a card - the B300 DataCrunch sells as "B300 SXM6 268GB". It is the dearest B300 we read today, above Oracle ($15.00) and Hyperstack ($7.40), as AWS is for the B200 ($14.24). Spot is unchanged: it needs signed API calls and is still not recorded. The definition of the published figure did not change. | unchanged | |
| 1.25.10 | patch | Coverage: Bentaus is read - AMD Instinct MI325X at $2.25 per GPU-hour on demand and $1.95 reserved, the two priced tiers of its card page. The MI325X now has three sources, beside DigitalOcean ($3.80) and TensorWave (from $2.25). Only AMD cards are mapped, as at TensorWave, so nothing Bentaus publishes can reach the H100 SXM Price Index. Its NVIDIA B300 page prices both tiers as a band ("$6.50 - $8.50 per GPU-hour") and is not read at all: one end of a band is not a rate. The reserved tier states no term, so the quote carries none. The definition of the published figure did not change. | unchanged | |
| 1.25.9 | patch | Coverage: Thunder Compute is read - RTX A6000 $0.35, L40 $0.79, A100 80GB $1.09 and H100 PCIe $3.20 an hour, each an instance at its smallest configuration (more CPU cores are an add-on), from its single-GPU price list; the 2- to 8-GPU tabs are exact multiples and are not read twice. Its H100 is the PCIe card, so it does not enter the H100 SXM Price Index. The definition of the published figure did not change. | unchanged | |
| 1.25.8 | patch | Coverage: fal is read - B300 $8.50, B200 $6.25, H200 $4.50, H100 $4.50 and RTX PRO 6000 $2.99 per GPU-hour, its serverless list price, each machine type with its CPU and memory. The "as low as" figures beside them are offered to custom deployments through sales and are not recorded. Its H100 names no form factor on the priced page, so it is the neutral H100 card and stays out of the H100 SXM Price Index. The definition of the published figure did not change. | unchanged | |
| 1.25.7 | patch | Coverage: Replicate is read - A100 80GB $5.04, H100 $5.49, L40S $3.51 and T4 $0.81 an hour, each an instance with its CPU and memory, from the self-serve hardware table. Its "Additional hardware" (H200, and 4- and 8-GPU machines) is sold by committed-spend contract and is not recorded. Its H100 names no form factor, so it is the neutral H100 card and stays out of the H100 SXM Price Index. The definition of the published figure did not change. | unchanged | |
| 1.25.6 | patch | Coverage: Azure's T4 (NCasT4_v3, $0.526 on demand) and A10 (NVadsA10_v5 from the 36-vCPU size up, the smallest carrying a whole card; the 1/6 to 1/2 sizes are fractions of a GPU and are not read), on demand, spot and 1- and 3-year reservations. Azure quotes are now never taken from an AT&T edge zone: the cheapest T4 was $0.343 in "ATT Dallas 1", an operator zone a customer does not simply deploy into, against $0.526 in every public US region. No quote published before came from one, so no figure changed. T4 now has six providers, A10 five. The definition of the published figure did not change. | unchanged | |
| 1.25.5 | patch | Coverage: the hyperscalers' inference cards. AWS - T4 (g4dn.xlarge), L4 (g6.xlarge), V100 (p3.2xlarge). Oracle - A10, V100, A100 40GB and RTX PRO 6000; its price list names three of them by an internal series, each matched to its card through the part number the price-list page prints beside the shape. Google Cloud - T4, V100 and L4 on demand and on 1- and 3-year commitments (the L4 was in the parser's map and never read); no spot rate for these three, because the page prints none and the catalog's T4 spot rate, $0.0315, fails the parser's plausibility floor. T4 now has five providers, A10 four. None of these is an H100, so the Price Index is untouched; the definition of the published figure did not change. | unchanged | |
| 1.25.4 | patch | Coverage: six cards the parser already read every hour and dropped as unmapped join the catalogue - NVIDIA A10 (Lambda, Modal, Cerebrium), T4 (Modal, Cerebrium, Baseten), A30 (E2E, Jarvislabs, Massed Compute, Seeweb), A40 (Denvr, E2E, RunPod), RTX A5000 (Massed Compute, Paperspace, RunPod, Vast.ai) and RTX A4000 (Hyperstack, Paperspace, Vast.ai). Two rates the RunPod API returns are not published because no RunPod page prints them: RTX A4000 (printed on its model page, not on /pricing, where every RunPod row links) and MI350X (no page at all). Baseten's A10G stays out: it is AWS's variant of the A10, not the A10. None of the six is an H100, so the Price Index is untouched; the definition of the published figure did not change. | unchanged | |
| 1.25.3 | patch | Coverage: TensorWave, an AMD-only cloud, is read - MI300X $1.71, MI325X $2.25, MI355X $2.95 per GPU-hour, each the "Starting at" figure its accelerator page prints for a whole node, so each is tagged "from" (advertised_from) like Sesterce and GMI Cloud. MI325X now has two providers, MI355X three. Voltage Park's and Latitude.sh's "Starting At" is still not recorded: that rule is about an H100 floor entering the Price Index, and TensorWave prices AMD only, which the parser enforces by mapping AMD ids alone. The definition of the published figure did not change. | unchanged | |
| 1.25.2 | patch | Coverage and one panel change. A new card, "NVIDIA H100 80GB (form factor not stated)", for sources that price an H100 without saying SXM or PCIe: GMI Cloud, Baseten, Cerebrium, E2E, Koyeb, Massed Compute, Northflank and Sesterce (whose row reads "SXM5 / PCIe"). Seven of them were read before and dropped; GMI Cloud was published as H100 SXM on the strength of a research note, not its page, which reads "NVIDIA H100 GPU". GMI Cloud therefore leaves the H100 SXM Price Index panel from the week of 2026-09-14; closed weeks keep its quote, and the base-100 index moves only on price changes of providers present in both weeks. The definition of the published figure did not change. | unchanged | |
| 1.25.1 | patch | Coverage: AMD cards on more price lists. MI300X is read at Azure (ND MI300X v5, $48.00 for an 8-GPU VM, $6.00 per GPU-hour, with its spot and 1- and 3-year rates), Oracle ($6.00 per GPU-hour) and Hot Aisle, an AMD-only cloud read for the first time ($2.99 per GPU-hour for its pay-as-you-go VMs; its bare-metal node, a one-month commitment priced above the VM, is not read). MI355X is tracked from today, priced by Oracle ($8.60) and by DigitalOcean spot ($4.50), whose spot cards are now read for the AMD cards; a spot card printed at the on-demand price (B300) is not published. The definition of the published figure did not change. | unchanged | |
| 1.25.0 | minor | The weekly Price Index series carries the middle half of each week: p25_usd_per_gpu_hour and p75_usd_per_gpu_hour in series[] of price-index.json, two new last columns of price-index.csv and a column of the weekly table on /price-index - the 25th and 75th percentile of the quotes in the week's receipt, the same quotes its median was taken over. Until now only the current week had it. By the owner's decision it was computed for every published week, closed weeks included, from the receipts those weeks already published (week of 2026-09-07: $3.12-$6.00; 2026-09-14: $2.89-$5.95). No published median moved; before adding the band, each week's receipt was checked to reproduce its published median. | unchanged | |
| 1.24.3 | patch | Coverage: the first AMD cards. AMD Instinct MI300X is read at Crusoe ($3.45), DigitalOcean GPU Droplets ($2.59, $1.91 for 12 months), RunPod Secure Cloud ($2.39) and Seeweb (EUR 2.35); MI325X at DigitalOcean ($3.80); MI350X at DigitalOcean ($6.16) and CloudRift ($3.65) - per GPU-hour, on-demand, each read from a page the parsers already opened and until now dropped as unmapped. RunPod's API also returns $0.50 for the MI300X on Community Cloud; no RunPod page prints that figure (the card page reads "from $2.39/hr"), so it is not published. MI355X is not tracked: its one source says "Contact sales". The Price Index is an H100 series and is untouched. The definition of the published figure did not change. | unchanged | |
| 1.24.2 | patch | Coverage: Vast.ai V100 16 GB is read - the marketplace lists 16 GB and 32 GB cards under one name, "Tesla V100", so only offers below 24 GB count. On this date the single-GPU floor was withheld (two verified offers, a floor needs three) and the multi-GPU quote was published ($0.48 for a 4-GPU machine, $0.12 per GPU). Marketplace floors stay out of every median. The definition of the published figure did not change. | unchanged | |
| 1.24.1 | patch | Coverage: Seeweb is read for the first time - listed in euro per GPU, converted at the ECB rate like Scaleway and OVHcloud: H100 SXM EUR 1.89, H200 EUR 2.60, A100 80GB EUR 0.99, L40S EUR 0.85, RTX PRO 6000 EUR 1.25, RTX A6000 EUR 0.74, L4 EUR 0.38 per GPU-hour on-demand, each with its 12-month rate. That the figure is per GPU was read from the page's own price script, which multiplies the rate by the GPU count. Spot is not read: the page promises up to 40% off while its script charges 40% of the price. EUR listings stay out of the Price Index. The definition of the published figure did not change. | unchanged | |
| 1.24.0 | minor | The risk tag the rate tables print is now documented and published as data: /methodology#risk says what it measures (whether capacity can be taken back mid-run, read off the market: spot and community high, serverless medium, on-demand and reserved low) and what it does not (provider reliability, stock, the cost of a commitment billed for idle hours), and prices.json and prices.csv carry it as risk. A new field; no published figure changed. | unchanged | |
| 1.23.2 | patch | Coverage: Northflank, Cerebrium and Baseten are read for the first time. Northflank: GPU-only per hour - L4 $0.80, A100 40GB and 80GB, RTX PRO 6000. Cerebrium: serverless per second, GPU only - L4 $0.80 per hour, A100 40GB and 80GB, L40S, RTX PRO 6000, H200, B200. Baseten: dedicated deployments per minute, instance - L4 $0.85 per hour, A100 80GB, B200. The L4 goes from seven rate cards to ten. All three list an H100 without its form factor, which is not recorded. Exoscale is recorded as not listed: its price file gives the same figures under EUR, CHF and USD. The definition of the published figure did not change. | unchanged | |
| 1.23.1 | patch | Coverage: Vast.ai lists the A100 under two names, "A100 SXM4" and "A100 PCIE"; both are now read as one book per memory size, as other providers' SXM and PCIe A100 already count as one card. The A100 80GB floor, withheld that morning because only two verified 80 GB SXM4 offers were rentable, is published again ($1.00 per GPU-hour single, on offers filtered to 80 GB by the marketplace). The definition of the published figure did not change. | unchanged | |
| 1.23.0 | minor | Beside the median rate card, from four rate cards: the middle half - the 25th and 75th percentile of the same rate cards, interpolated as percentile_cont, so the 50th is the median itself - on /spread, each card page, gpus.json (with the median and the rate-card count, new there too) and, for the index quotes, price-index.json. /spread now takes its rates from the same rate cards as every other card median: until this date it kept its own filter and counted community capacity (RunPod Community Cloud) as on-demand, which 1.18.0 had already removed everywhere else, so its cheapest rate and median could differ from the card pages. No other published figure changed. | changed | |
| 1.22.1 | patch | Coverage: the Vast.ai floor for the RTX PRO 6000 is read for the first time ($1.34 per GPU-hour single, $1.33 per GPU on an 8-GPU machine). The parser had asked the marketplace for "RTX PRO 6000", a name it does not use; it lists the card as "RTX PRO 6000 S" and "RTX PRO 6000 WS", read now as one book, while the lower-power Max-Q edition stays out. A check of every name the parser reads the same day found no other card sold under one name in two memory sizes besides the A100 and RTX 4090 already separated that morning. The definition of the published figure did not change. | unchanged | |
| 1.22.0 | minor | A new file: /changelog.xml, an Atom feed of providers' own rate changes - the newest 100 moves from changelog.json, one entry per move with a stable id, linked to the card page and the source. The marketplace floor (Vast.ai) is left out of the feed because it moves almost every hour; on this date the newest 100 moves were all Vast.ai floors. No published figure changed. | unchanged | |
| 1.21.3 | patch | Coverage: E2E Networks and Koyeb are read for the first time. E2E: on-demand per GPU-hour, before taxes - L4 $0.57, L40S, A100 80GB and 40GB, RTX PRO 6000, H200 and B200. Koyeb: serverless GPU instances with CPU, memory and disk included - L4 $0.70, RTX A6000, L40S, A100 80GB, RTX PRO 6000, H200 and B200. The L4 goes from five rate cards to seven. Both list an H100 without its form factor, which is not recorded. The definition of the published figure did not change. | unchanged | |
| 1.21.2 | patch | Coverage: Atlantic.Net and Denvr Dataworks are read for the first time. Atlantic.Net: H100 NVL ($3.94 per GPU-hour) and L40S ($1.67) servers, on-demand and on 1- and 3-year terms, the hourly figure as the page prints it (monthly price / 730). Denvr: on-demand GH200 ($3.87), H100 SXM ($2.45), A100 80GB ($1.45) and A100 40GB ($1.15) per GPU-hour. The GH200 gets a third rate card and the H100 NVL a fourth; Denvr's H100 SXM rate enters the Price Index from the next hourly scrape. The definition of the published figure did not change. | unchanged | |
| 1.21.1 | patch | Coverage: two Azure series the Retail Prices API already returned are read - NC H100 v5, one or two H100 NVL (94 GB) per VM ($6.98 per GPU-hour on-demand in eastus, spot and 1- and 3-year reservations), and NCv3, one to four V100 16 GB per VM ($3.06 per GPU-hour). The H100 NVL, until now on two providers' rate cards, gets a third. The definition of the published figure did not change. | unchanged | |
| 1.21.0 | minor | price-index.json carries the typical discount of /price-index#discounts as market_discounts: per term (spot, 1-year, 3-year) its status, median, range, pairs and providers, and every rate pair behind it with both rates, the difference in $ and % and the source link - withheld terms included, with their figures null and the reason. New fields; no published figure changed. | unchanged | |
| 1.20.0 | minor | A new aggregate figure on /price-index: the typical discount of spot and committed rates against the same provider's on-demand rate for the same card, region and price scope - the median of those per-quote differences across every card, with the number of rate pairs and providers behind it, and a term stated only with three or more providers. On this date: spot -50% (33 pairs, 5 providers), 1-year commitments -31% (9 pairs, 3 providers); 3-year commitments withheld, published by Azure and Google Cloud only. Rates with no stated term are not one term and are left out. The Price Index is unchanged. | unchanged | |
| 1.19.0 | minor | Two new cuts of the same quotes on card and provider pages. Every reserved and spot rate shows its difference, in dollars and per cent, from the same provider's on-demand rate for the same card, region and price scope (Azure H100 SXM 3-year $4.86: -$6.21, -56% against its $11.06); a rate with no such on-demand rate beside it shows none. Card pages state the cheapest and dearest on-demand quote priced as a whole 8-GPU node (H100 SXM: $29.23 Scaleway to $88.49 Azure per node-hour), and eight GPUs at the cheapest on-demand rate beside it. No published figure changed. | unchanged | |
| 1.18.1 | patch | Coverage: CloudRift and Beam are read for the first time. CloudRift: on-demand RTX 4090 ($0.39), RTX 5090 ($0.60), A100 80GB ($1.05) and V100 per GPU-hour, and a 1-month reserved rate for those and for L40S, RTX PRO 6000 and H200, which it sells reserved only. Beam: on-demand machines "from" a price (RTX 4090 $0.44, RTX 5090 $0.72, and B200, H200, H100 PCIe, A100 80GB, RTX PRO 6000, L40S, RTX A6000) and serverless GPU seconds for RTX 4090, RTX 5090 and H100 PCIe. The RTX 4090 and RTX 5090, until now on one provider's rate card, get a median rate card. Neither lists the H100 SXM, so the Price Index does not move. The definition of the published figure did not change. | unchanged | |
| 1.18.0 | minor | The median rate card takes on-demand rates proper, as "cheapest on-demand" has since 1.17.0: community capacity (RunPod Community Cloud) leaves it with marketplace floors, so the median and the count of providers listing lower on a provider page rest on the same providers. Where a provider had only community capacity for a card it drops out of that card's median. Moved on this date: A100 40GB $2.10 across 9 providers to $2.42 across 8; V100 $0.51 across 4 to $0.79 across 3; H100 NVL $2.75 to $3.05; B300 $7.73 to $7.85; B200 $6.69 to $6.79; RTX PRO 6000 $1.87 to $1.99; H200 $4.50 to $4.54; RTX 6000 Ada $0.94 to $0.96. The cheapest rate card moves with it (RTX 4090 $0.34 to $0.74, RTX 5090 $0.69 to $0.99, both RunPod Secure Cloud), and so does break-even at the median. H100 SXM ($3.90), A100 80GB, L40S and the Price Index are unchanged. | changed | |
| 1.17.0 | minor | "Cheapest on-demand" and "who is cheaper" rest on on-demand rates proper: community capacity (RunPod Community Cloud, on third-party hosts) and marketplace floors (Vast.ai) are listed as on-demand at the source but are shown in their own column on /cheapest and /compare, never decide a verdict on /compare/{a}-vs-{b}, and are no longer the cheapest rate on /buy-vs-rent. Card pages already labelled them community, high risk; the Lambda vs RunPod B200 verdict had been decided by a Community Cloud rate. Pair pages stay; a shared GPU resting on community capacity is shown and not counted. The median rate card and the Price Index are unchanged. | changed | |
| 1.16.0 | minor | Every retail price says what kind of price it is - a store's own price, a marketplace seller's listing, or a completed sale - and a retail median is computed inside one kind, never across. A card leads with the store median where three or more stores carry it, else the marketplace median, each named, with the other kind beside it. Three medians changed: A100 80GB $15,760 over 14 sellers of both kinds became $11,986 at 7 stores and $16,999 at 7 marketplace sellers; RTX 4090 $3,195 became $3,103 and $3,200; RTX 5090 $4,841 became $4,753 and $5,250. Break-even on /buy-vs-rent and the card pages moves with them. Rental figures, prices.json and the Price Index are untouched. | changed | |
| 1.15.0 | minor | A median rate card is published only with two or more providers. A card on one provider's rate card (RTX 4090 and RTX 5090 on this date, where most supply is the Vast.ai floor) shows that rate card by name and the cheapest marketplace floor beside it, and gets no break-even, no median column on /buy-vs-rent and no median comparison on /gpus - a median of one is a provider's price, not a market figure. | unchanged | |
| 1.14.0 | minor | One card median across the site, the median rate card - each provider once, at its cheapest on-demand rate, marketplace floors out - printed on card pages, /buy-vs-rent, the GPU comparisons and their JSON-LD, as the cards table and /spread already did. It replaces a median over every on-demand listing, which counted a provider once per region or tier and took marketplace floors in (H100 SXM: $3.85 to $3.90). The Price Index is unchanged. | changed | |
| 1.13.0 | minor | The weighting rule is stated where the median is printed - one quote per provider and region, so a provider that lists several regions enters once per region - and every week carries a check next to the index: the same quotes as the median of each provider's own median, with the largest provider's quote count and an HHI, in price-index.json as weighting_check and on /price-index. The index keeps its definition; no published figure changed. | unchanged | |
| 1.12.0 | minor | Every live quote carries its change over seven days: the listed figure per GPU against the same quote's read seven days before the newest scrape (within 48 hours of that moment), from our own reads only, in prices.json and its CSV as listed_7d_ago, read_7d_ago_at and change_7d_pct. A card's change is the geometric mean of its rate cards' own changes. New fields; no published figure changed. | unchanged | |
| 1.11.0 | minor | Vast.ai multi-GPU machines are read as a second quote per GPU model, region "Multi-GPU machines": the machine's own price with its GPU count, the per-GPU figure that price divided by the count, from the machine whose per-GPU ask is the median of the five cheapest, under the same guards. The single-GPU floor did not change and never mixes with it, because a share of a multi-GPU machine is not a machine you can rent alone. | unchanged | |
| 1.10.2 | patch | Coverage: Massed Compute is read for the first time - on-demand H100 SXM ($2.89), H100 NVL, A100 80GB, L40S, L40, RTX PRO 6000, RTX 6000 Ada and RTX A6000 from its VM pricing page. Headings that name no form factor (a bare H100 80GB) or a card we do not track as such (H200 NVL) are left out rather than guessed. Its H100 SXM rate enters the Price Index from the next hourly scrape. The definition of the published figure did not change. | unchanged | |
| 1.10.1 | patch | Coverage: Crusoe is read for the first time - on-demand H100 SXM ($3.90), H200, A100 80GB and L40S from its pricing table; GB200 and B200 are contact-sales there and are not recorded. Its H100 rate enters the Price Index from the next hourly scrape, as any provider's would. The definition of the published figure did not change. | unchanged | |
| 1.10.0 | minor | A marketplace floor withheld as too thin is recorded with the read that withheld it, and the quote is marked thin_market from that read on: it keeps its last floor with that floor's read time, leaves every computed figure at once instead of counting for up to 48 hours, and never ages into unconfirmed or stopped while the source is still being read. prices.json and its CSV gain withheld_at and withheld_reason. Which quotes count as current changed; no published definition did. | unchanged | |
| 1.9.0 | minor | The Price Index is also published cut by operator class: the same weekly median over the same quotes, split into hyperscalers (AWS, Azure, Google Cloud, Oracle) and specialist GPU clouds, each with its quotes and providers - by_operator_class in price-index.json, four columns in price-index.csv, a table on /price-index. Marketplaces have no value, because their floors never enter the index. Every provider carries its class (providers.operator_class, filled by hand). A new cut, not a new definition: the published value did not move. | unchanged | |
| 1.8.0 | minor | Each weekly receipt of the Price Index lists every H100 SXM on-demand quote read for that week, not only the ones that made the value: the quotes kept out carry included=false and, in a new excluded_by column, the rule that kept them out - marketplace_floor, eur_listing or not_confirmed_48h, the codes price-index.json publishes as exclusion_rules. The reproduce commands read the included rows, so no published figure moved. | unchanged | |
| 1.7.0 | minor | The home and card pages print one modelled figure for the first time: cost per million output tokens, a listed rate divided by a throughput NVIDIA published for that card, with the run that produced it (model, framework, precision, sequence, concurrency, date) beside it and a link to the benchmark. It is published per card and never as a ranking, because NVIDIA runs each card in its own configuration. Every other figure on the site is still read from a provider page. | unchanged | |
| 1.6.1 | patch | Coverage: the committed-use rates of Google Cloud and Azure are read for the first time, one-year and three-year, from the same endpoints as their on-demand prices. Azure publishes a reservation as the total for its term, so the hourly figure is that total divided by the hours in the term (8,760 per year). AWS still publishes only on-demand where we read it. No existing figure changed. | unchanged | |
| 1.6.0 | minor | Every quote says which market it is from: on-demand, spot, community (shared capacity - RunPod Community Cloud and the Vast.ai floor), serverless (billed by the second, no instance to rent - Modal), or a commitment with a stated term (Paperspace, one year) or without one (Nebius, Hyperstack, Together). Read from each source, never inferred from the price; Together was checked against its own page and sells on-demand clusters, not serverless. The Price Index still selects exactly as before, so no published figure moved. | unchanged | |
| 1.5.0 | minor | Every quote carries the name the provider gives what was priced - instance type, SKU, machine family or marketplace offer id (p5.48xlarge, HGX B200, offer 48635744) - in prices.json, its slices and prices.csv. Fifteen of twenty-one sources publish one; the rest list a single rate per card and the field stays empty rather than being derived. No published figure changed. | unchanged | |
| 1.4.0 | minor | Every published figure carries methodology_version: top level in price-index.json and prices.json and their slices, a column in prices.csv, price-index.csv, price-index-history.csv and in each weekly receipt. This dated log is published beside it. | unchanged | |
| 1.3.2 | patch | A Vast.ai marketplace floor is withheld when the window it came from cannot carry one: fewer than three verified single-GPU offers, or a median more than 3x the cheapest ask beside it. On 2026-09-11 that window published $53.34 per RTX 5090 GPU-hour. Marketplace floors never enter the index, so no published index week moved. | unchanged | |
| 1.3.1 | patch | Collection and coverage only: RunPod spot rows that repeated the on-demand price are no longer published (the API returned spot == on-demand for every GPU and tier), and Sesterce was added as the 21st provider. The definition of the published figure did not change. | unchanged | |
| 1.3.0 | minor | A quote not read again within 48 hours of the newest scrape stops counting as current: it is marked unconfirmed (stopped after 7 days), leaves every computed figure including the running index week, and carries last_confirmed_at and status in the files. Each published week ships its receipt, every live row says since when the figure has been listed, and a withheld week prints why. | unchanged | |
| 1.2.0 | minor | A chain-linked base-100 index is published beside the dollar median, so a change can be read without the panel moving it, and the reconstructed SkyPilot-catalog history is published as a separate, second-hand series that is never merged with our own. | unchanged | |
| 1.1.0 | minor | A week is published only when at least 8 providers contributed; weeks below that are withheld, never backfilled. Every quote declares its price_scope, marketplace floors and quotes listed in a currency other than USD are shown on the site but kept out of the index, with the listed price and the ECB rate beside the converted figure. | unchanged | |
| 1.0.0 | base | First published week of our own series: the Cloud GPU Price Index is the median across providers of the latest publicly listed on-demand NVIDIA H100 SXM rate, per GPU per hour, node prices divided by GPU count, each quote linked to the page it was read from. | unchanged |
The log starts on , the first published week of our own series; the entries up to November 4, 2026 were reconstructed from the repository while that history was six days long, and every change from here is recorded as it ships. The reconstructed history from the SkyPilot catalog is a separate, second-hand series and carries its own description on the index page — it is not versioned here.
Cite this page
Published under CC BY 4.0: use and redistribute with attribution and a link. Cite the linked source for a price itself, and cite us for the normalization, the index and the comparison. Cite a figure together with its methodology_version; version 1.33.64 is what the files carry today.
| Plain | GPU Economy, "Methodology changelog", November 4, 2026. https://gpueconomy.com/methodology/changelog |
|---|---|
| Markdown | [GPU Economy, "Methodology changelog" (November 4, 2026)](https://gpueconomy.com/methodology/changelog) |
| APA | GPU Economy. (2026-11-04). Methodology changelog. Retrieved from https://gpueconomy.com/methodology/changelog |
Data updated: