GPU Price Comparison
Compare H100, H200, A100, B200 and RTX hourly prices per model across decentralized and crypto-payable networks.
runs 100% locallyno third-party requests
Pick a GPU and see every compute provider that offers it. The $/GPU-hr column is the price for that model from our sourced rate card (checked 2026-09-24, or the live Akash index); the cheapest SKU across a provider's whole catalogue is a separate column so one number is never passed off as another. Unknown means not published.
pick a GPU…
Compare the whole cost, not just the hour
GPU-hour prices hide the extras: egress/bandwidth, storage, cold-start time and host reliability. Decentralized markets can be 2-3x cheaper than hyperscalers for the same GPU, but a % of community hosts drop mid-job - so for anything long-running, weight reliability as heavily as price. Every rate row names its source and check date. Live rates: GPU price index.
Questions about the price index
What is a per-model price?
Only listings that name the model count. The cheapest any-SKU price of a provider used to be reported as an H100 price, which flattered heterogeneous markets.
Why do some providers have no price?
Because they do not publish that model or it is not in the normalised index. An unlisted model is shown as unknown, not estimated.
Which access options are shown?
Rental paths only - on-demand, spot and preemptible where documented. Bare metal and colocation are different products.