GRIDCOMPUTE

Developers

Free API and price badges.

Hourly GPU rental prices and grid power data as JSON and CSV. No key, no sign-up, CORS open. Free to use under CC BY 4.0: credit “GridCompute GPU Price Index” with a link to gridcompute.ai.

Just want to know when a price moves? Set an email price alert on any GPU page. No account needed.

Embed a live price badge

An image that updates by itself. Works in READMEs, blogs and newsletters, with no script.

H100 SXM rental price on gridcompute.ai
HTML

      
Markdown

    

The badge shows the median Vast.ai asking price this hour. GPUs not on the marketplace show the lowest published rate card, marked “from”. The history chart plots the hourly median with the middle half of listings as a band, from stored observations only (nothing before tracking began is filled in). Images refresh about every 15 minutes.

API reference

Base URL https://gridcompute.ai. All endpoints are GET. Prices are in USD per GPU-hour, asking prices, not offers. Timestamps are Unix seconds at the start of the UTC hour. Missing values are null, never estimated.

Data updates once an hour, a few minutes past the hour, and responses are cached for 5 minutes. There is no hard rate limit; please cache on your side rather than polling more than hourly. New fields may be added; existing fields will not be renamed or removed without notice on this page.

Market prices come from public Vast.ai listings. Rate cards: RunPod (Secure Cloud on-demand), Verda (on-demand, 1-GPU instance), Azure (pay-as-you-go Linux VM, East US 2, per GPU). Full method on the homepage.

GET /api/latest

Every tracked GPU: the current marketplace snapshot, each rate card, 24h and 7d change, and the last 7 days of hourly medians.

meta.first, meta.last
Unix seconds (UTC) of the first and latest hourly snapshot
meta.snapshots
Number of hourly snapshots stored
lists[]
The rate cards tracked: key, name, note
gpus[].key, name, mem, tier
GPU id, display name, memory in GB, family
gpus[].market
Vast.ai marketplace asking prices this hour: offers, gpus, min, p25, median, p75. null when not listed this hour
gpus[].lists
Rate card price per GPU-hour this hour, keyed by rate card. null when not published
gpus[].change24h, change7d
Fractional change of the market median (0.012 = +1.2%). null until enough history exists
gpus[].spark
[ts, median] pairs for the last 7 days
gpus[].list
Deprecated alias of lists.runpod
GET /api/history?gpu=h100-sxm&days=30

One GPU over time. Hourly rows up to 21 days, daily averages beyond that (days 1–365, default 30).

resolution
hourly or daily average
vast[]
Objects with ts, min, p25, median, p75, offers
runpod, verda, azure
[ts, price] pairs for each rate card
GET /api/export.csv?gpu=all&days=90

Raw stored rows as CSV. gpu is a GPU id or all.

columns
timestamp_utc, gpu, source, offers, gpus, min, p25, median, p75. Rate card rows repeat the price in every stat column and leave offers and gpus empty
GET /api/gpus

The tracked GPU ids, names, memory and family.

GET /api/power/latest

Each grid region: latest hourly price and load, 24h average and range, and 7-day series.

regions[].price
ts, value, avg24h, high24h, low24h, in the region currency per MWh. null where no price is published
regions[].load
ts, value, peak24h in MW
GET /api/power/history?region=ercot&days=30

Hourly price and load for one region (ercot, pjm, ontario, quebec).

GET /api/power/export.csv?region=all&days=90

Raw hourly power rows as CSV.

GET /api/spread/latest?gpu=h100-sxm&pue=1.3

Compute spark spread on every grid for one GPU: latest matched hour of rental minus power cost. grid (optional) limits it to one region; pue 1.0–2.0, default 1.3. GPUs: those with board power on file (H100 SXM, B200, A100 SXM, L40S, RTX 4090, RTX 5090).

as_of
Unix seconds of the latest matched hour across grids
board_kw, board_kw_source, pue
Inputs: GPU board power from the spec sheet linked, and the PUE used
method, sources[]
Formula in words, and each input source with its licence
grids[].unavailable
Why a grid has no spread (Québec: no wholesale market; PJM: price not redistributable), else null
grids[].rental, rental_source
USD per GPU-hour: Vast.ai median (vast), else the lowest rate card
grids[].power_local_per_mwh, currency, fx
Wholesale price as published (ERCOT USD, Ontario CAD) and, for CAD, the Bank of Canada usdcad rate and its date
grids[].power_kw, power_usd_per_mwh, power_cost_hr
Facility draw (board kW × PUE), price in USD, and power cost per GPU-hour
grids[].spread, margin
rental − power_cost_hr and spread ÷ rental (can exceed 1 when power prices are negative)
GET /api/spread/history?grid=ercot&gpu=h100-sxm&days=7

Hourly spread for one grid and GPU (days up to 90). Same fields as above, per hour in hours[]; hours missing rental or power are left out, never filled.

GET /api/basis/latest

Futures basis: the latest benchmark level per GPU, provider and tier, against GridCompute’s market median and lowest rate card at that hour. Empty until a benchmark publisher permits display; levels stay under the publisher’s terms, not CC BY.

levels[].benchmark
value, source_url, method (manual or collected)
levels[].market, ratecard
GridCompute value with ts, source, method
levels[].basis_market, basis_pct_market
GCX median − benchmark, and that ÷ benchmark
levels[].basis_ratecard, basis_pct_ratecard
Lowest rate card − benchmark, and that ÷ benchmark
GET /api/basis/history?gpu=h100-sxm&days=30

Every benchmark level for one GPU in the window, each compared with the hourly observation at its time. Tiers are never blended.

GET /api/health

Time and outcome of the last collection run. ok is false if no run in the last 3 hours.