Put the numbers to work
Three ways to use the dataset: price your own job, decide whether to rent GPUs or buy tokens, or take the whole thing as JSON and build something else.
What will this job cost?
GPU count × hours × today’s median, with the arithmetic shown. It also prices the same job at the cheapest and dearest tracked providers, because that gap is usually larger than people expect.
e.g. 8 × NVIDIA H100 SXM for 72h = $2263.68 at the median · $1146.24–$6725.09 across providers
Where’s the break-even?
Renting a GPU is a fixed hourly cost; an API is a variable per-token one. Below some daily volume the API wins, above it the GPU does. You supply the throughput — we won’t invent a tokens-per-second number for your model.
Take the data
Every number on this site is available as JSON — current prices, full history, the Index, model prices, events, and the spot-vs-on-demand discount. 1,000 requests per day per IP, no key required, attribution requested. There are also embeddable SVG badges that stay live wherever you paste them.