prism.

Product / Pricing

GPU compute at $0.80 per hour, and what it cost us.

Per-second billing with a five-minute quote, a defined maximum escrow amount, and automatic return of unused funds after settlement. The index below is drawn from real purchases and onchain settlements rather than list prices.

01

Current rate

Displayed rate
$0.80 / hr
Exact rate
0.7992 USDG
Metering
Per second

An offer charges 222 USDG base units per second, equal to 0.000222 USDG per second or 0.7992 USDG per hour, which the interface rounds to $0.80. The rate is the same across every GPU class the network currently serves.

02

What compute actually costs

There is no public print of what an hour of a datacenter GPU trades for. Prices are quotes on landing pages, and what anyone paid stays private. Because every lease here settles onchain, this network produces the record as a side effect of operating, so it is published rather than kept.

Loading the current index.

The gap between the two columns is the margin the network runs on, and it is visible for the same reason everything else is.

03

Service limits

  • Quotes expire after five minutes and do not reserve capacity.
  • A lease may run for at most six funded hours.
  • Maximum escrow is capped at 50 USDG per lease.
  • Billing begins only after GPU, pricing, and access-readiness checks pass.
  • Unused escrow is returned when settlement is finalized.
  • Wallet and Robinhood Chain gas fees are separate.
04

Confidential inference

Renting a machine by the hour is one way to buy compute here. The other is a single generation, priced per call and paid from a wallet with no account and no API key. Nine models run at the confidential grade, among them DeepSeek, Llama 3.3 70B, GPT-OSS, Qwen and GLM.

A call costs a fixed charge plus a rate over the output cap it asks for, quoted in full before payment and settled in USDG. Today that lands between 0.011 and 0.025 USDG for a full 1,024-token answer, depending on the model. The live model list carries the current rate for each one.

You pay for the cap you request rather than the tokens produced, so a request that asks for less costs less. An answer whose attestation does not verify is withheld, and that call is still charged.

05

Provider economics

Finalized usage allocates 90% of the confirmed charge to the provider and 10% to the Prism as the service fee. The funded maximum is not the final charge; settlement is based on confirmed runtime, subject to contract limits and the dispute process.

Pricing is quote-based and may change with capacity and market conditions. Always review the exact rate, duration, escrow amount, chain, and contract before signing a funding transaction.