Machine-payable compute intelligence for AI agents.
Answers one question for AI agents and the engineers who run them: which compute can actually run this job right now for the best price. Official vendor pages only.
x402 · $0.005 USDC on Base or Solana per call. No key.
See a free example response →Inference rows507
GPU rows96
Prices read2026-10-07 04:18:37Z
Book statefresh
Entire fleet
One live registry, every company we price. The fleet currently spans 16 companies across 11 inference feeds and 6 GPU feeds. Each feed links to the official source it reads.
16 companies
11 inference feeds
6 GPU feeds
Anthropic
anthropicGroq
groqTogether AI
togetherDeepSeek
deepseekMistral AI
mistralFireworks AI
fireworksxAI
xaiRunPod
runpodLambda
lambdaCrusoe
crusoeMiniMax
minimaxCerebras
cerebrasNebius
nebiusHyperstack
hyperstackHow to call it
One HTTP request. Describe the job, get back the cheapest compute that can run it, with the arithmetic and the source page behind the number.
In four steps
- Send your job: either tokens (
work=inference) or GPU hours (work=gpu). - You get a
402back with a payment challenge listing Base and Solana. - Sign for whichever chain you hold USDC on and resend with the payload in
PAYMENT-SIGNATURE. - The
200carries the answer, plus your receipt inPAYMENT-RESPONSE.
Two example calls
curl "https://compute-state-api.replit.app/v1/cheapest?work=inference\
&model_tier=cheap&est_input_tokens=2000\
&est_output_tokens=500"
curl "https://compute-state-api.replit.app/v1/cheapest?work=gpu\
&gpu_model=h100&gpu_hours=24"Want to see the exact response shape before paying anything?
GET /v1/example →Free vs paid
Everything you need to understand and integrate the API is free. The only thing you pay for is a live answer about your specific job.
| Endpoint | Cost | What you get |
|---|---|---|
| /v1/example | free | A fixed sample answer in exactly the paid shape, so you can build against it first. |
| /openapi.json | free | The full spec: parameters, response fields, error codes. |
| /.well-known/x402 | free | Where to pay and how much, in machine-readable form. |
| /llms.txt · /skill.md | free | Plain-text instructions so an agent can use this API without a human. |
| /healthz | free | Whether the price data is current enough to answer right now. |
| /privacy · /privacy.txt | free | What this service stores about a caller, for a person and for an agent. |
| /v1/cheapest | 0.005 USDC | The live answer for your job: cheapest option, the cost breakdown, alternatives, what was excluded and why. |
What comes back
Not just a number — a number you can explain to whoever asks why you picked it.
A price with a date
Every rate is read from the vendor's own pricing page, not an aggregator, and stamped with when it was read. Anything older than 15 minutes is left out of the answer.
The math, shown
The total is split into the parts it was added up from — each with the published rate, its unit, and whether that rate is list, effective, or unknown.
Returned with a reason
Every SKU that matched but cannot be used is returned with the reason — retired, unpublished price, waitlist, or a rate we will not invent.
When you are charged
No body, no settle; no settle, no charge. Your query and the freshness of the price data are checked before any payment is asked for. Your payment is verified and your answer built before any money moves.
| Status | What happened | Charged |
|---|---|---|
| 400 | Bad query. Before settlement. | no |
| 402 | Payment required or signature rejected. Before settlement. | no |
| 503 | No fresh row can answer this query. Before settlement. | no |
| 404 | Wrong path (this API has no such route). Before settlement. | no |
| 200 | Verified, ranked, then settled. Body computed before money moved; sent after. | yes |
| 404 | Paid no-match: verified, ranked, nothing in the pinned book fit. disqualified explains what came close. | yes |
| 500 | Process faulted before settlement. Operator fault. You were not charged. | no |
You are charged only when settlement succeeds and we have an answer in hand (200 or a no-match 404). 400, 402, unpaid 503, and 500 all happen before money moves. A 500 is our crash, not a stale book — we do not settle if we cannot build the body. If the HTTP response is lost after settlement, retry the same signature; we will not charge twice.
Payment
Pay per call with USDC over x402. No account, no API key, no subscription.
A listing probe of GET /v1/cheapest with no query string is a 402, not a 400. work is required to rank. It is not required to see the challenge.
| Field | Value |
|---|---|
| Protocol | x402 v2, scheme exact |
| Price | 0.005 USDC per call, both chains |
| Base | eip155:8453 |
| Solana | solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp |
| Facilitator | facilitator.payai.network |
Sources
Official vendor pages and APIs, nothing else. GPU coverage is on-demand list price only across RunPod, Lambda, Fireworks, Crusoe, Nebius and Hyperstack — no spot bids, no reserved or tailored commitments and no marketplace resellers — which is why the clouds here post a rate you can buy by the hour rather than a bid. Fireworks appears twice because it sells two products, serverless tokens and rented GPUs, on two separate pages.
| Provider | Work | Source |
|---|---|---|
| OpenAI | inference | platform.openai.com/docs/pricing |
| Anthropic | inference | docs.claude.com/en/docs/about-claude/pricing |
| inference | ai.google.dev/gemini-api/docs/pricing | |
| Groq | inference | console.groq.com/docs/models |
| Together AI | inference | www.together.ai/pricing |
| DeepSeek | inference | api-docs.deepseek.com/quick_start/pricing |
| Mistral AI | inference | mistral.ai/pricing/api/ |
| Fireworks AI | inference | docs.fireworks.ai/serverless/pricing |
| xAI | inference | docs.x.ai/developers/models |
| RunPod | gpu | www.runpod.io/pricing |
| Lambda | gpu | lambda.ai/service/gpu-cloud |
| Fireworks AI | gpu | fireworks.ai/pricing |
| Crusoe | gpu | www.crusoe.ai/cloud/pricing |
| MiniMax | inference | platform.minimax.io/docs/guides/pricing-paygo |
| Cerebras | inference | api.cerebras.ai/public/v1/models |
| Nebius | gpu | nebius.com/prices |
| Hyperstack | gpu | www.hyperstack.cloud/gpu-pricing |
State suite
Four APIs, one question each. This one answers what a job costs to run right now. It is the wrong tool for seat prices, uptime, or SDK breaks.
| API | Question it answers | |
|---|---|---|
| Pricing State | What is this vendor's seat / list ladder? | live |
| Status State | Is the vendor up? | live |
| Changelog State | Did the SDK ship a break? | live |
| Compute State | Which compute can actually run this job right now? | you are here |