Every price on Compute Screener is free to use: JSON files you can call like an API, and daily CSV files. No key and no sign-up. The data is CC BY 4.0: use it, including commercially, with credit to "Compute Screener (computescreener.com)".
JSON API
Plain GET requests to the files the website itself reads. They can be fetched from a browser on any site. GPU files are updated every 2 to 3 hours, token files once a day.
https://computescreener.com/data/| File | What it returns |
|---|---|
index.json | Latest GPU rental prices in USD per GPU-hour. Per GPU: value, date, chg_1d, chg_7d, chg_30d (fractions: 0.012 is +1.2%), its benchmark contract, and each provider's average in providers (name, avg standardized, posted, read, src). A GPU not yet published has value null. |
series.json | Each GPU's daily history: d (US Central day), v (value), final (false for today's provisional value), p (providers), w (weekly history, before 23 October 2026). |
providers.json | Each provider's offers for each GPU: the lowest and highest posted price per GPU-hour and the cheapest configuration. |
tokens.json | Latest AI token prices in USD per 1M tokens (3 input : 1 output) for each company, with each model's input, output, blended price and share of the company's tokens. |
token_series.json | Each company's daily token index: d (UTC day), v, final, h (history rebuilt from earlier posted prices). |
models.json | Each model in the token index: its daily input and output prices and share of its company's tokens. |
Examples
The H100 price, from a terminal:
curl -s https://computescreener.com/data/index.json | jq '.gpus.H100.value'Python, every GPU with its 1-day change:
import requests
idx = requests.get("https://computescreener.com/data/index.json").json()
for gpu, e in idx["gpus"].items():
if e["value"] is not None:
print(gpu, e["value"], f"{e['chg_1d']:+.1%}")JavaScript:
const idx = await (await fetch("https://computescreener.com/data/index.json")).json();
console.log(idx.gpus.B200.value);CSV
Daily CSV copies of the same data, refreshed every day on GitHub at renshuBTC/compute-screener. Each link below is the file itself:
| File | One row per |
|---|---|
gpu_index_daily.csv | GPU and day: the index in USD per GPU-hour |
gpu_providers_latest.csv | GPU and provider: the provider's latest average |
token_index_daily.csv | AI company and day: the token index in USD per 1M tokens |
token_models_latest.csv | Model: its latest input, output and blended price |
token_models_daily.csv | Model and day: its input and output prices |
In pandas:
import pandas as pd
df = pd.read_csv("https://raw.githubusercontent.com/renshuBTC/compute-screener/main/data/gpu_index_daily.csv")Good to know
- How every value is calculated: Methodology. Prices change every 2 to 3 hours, so polling more often adds nothing.
- In Claude, ChatGPT and other AI apps, the MCP connector answers with the same figures.
- Prices are information, not investment advice. Questions: Support.