Get the weekly radar
One short email with the freshest products, every week.
Grounded in available product and source data
Seventy-six providers publish GPU rental rates and none of them present those rates the same way. Compute Prices collects them alongside LLM inference costs and normalises both into one cloud GPU price comparison: 65 GPU providers across 67 GPU models on the compute side, 21 LLM providers across 300 models on the inference side.
The API is the part built for machines rather than readers. A free key returns pricing as JSON in one stable shape across providers, which removes the per-vendor scraping an agent would otherwise need to do. That free tier allows 750 requests a day and carries seven days of trend history; the Pro tier extends history to 90 days. Both include current rates and week-over-week movers, and the site surfaces a biggest-drops view rather than showing only what things cost today.
What the site is unusually direct about is the limits of its own numbers. Pricing is collected automatically from public sources and provider APIs, and the page states the figures may not reflect regional variations, minimum commitments, reservation discounts or negotiated enterprise pricing, directing readers to confirm on the provider's official page before purchasing. That caveat bites hardest at the enterprise end, where a list rate is rarely the rate anyone pays.
Teams sizing a training or serving budget get the most out of it, since compute and inference sit in the same view instead of two separate research tasks. The fit is thinner for anyone who has already negotiated committed-use terms with one provider, because the comparison is built from public list rates and has no way to see a private contract.
The site refreshes daily and displays how long ago the last run finished, which read under four hours at the time of writing. It also states the data is gathered automatically from public sources and provider APIs rather than entered by hand, so freshness depends on what those sources expose.
Yes. The site publishes a JSON API returning provider, GPU model, hourly price, pricing type and a last-updated timestamp, with a free key covering 750 requests a day. It positions the API as built for AI agents that need pricing without scraping each vendor separately.
Treat it as a shortlist step rather than a quote. Narrowing to two or three providers here and then confirming the current rate, the region and any minimum commitment with each one directly is the safer order, because the site lists reservation discounts and negotiated enterprise terms among the things its figures do not capture.
Longer history, according to the site: the free key carries seven days of price trends and Pro extends that to 90 days. Current rates and week-over-week movers are stated as available on both tiers, so the paid upgrade is about looking further back rather than seeing more providers.
