--- name: compute-state description: Find which compute can actually run a job right now — LLM inference or GPU hours — from official vendor prices, paid per call over x402. --- # Compute State Ask one question: Which compute can actually run this job right now? Base URL: https://compute-state-api.replit.app Brand mark: https://compute-state-api.replit.app/brand/hummingbird-logo.png No API key. No signup. ## What this is, and is not Compute State ranks current published list and effective rates from official vendor pages for the query you sent. It is a decision aid, not a quote, invoice, or promise the vendor will accept the job or bill that amount. Confirm capacity, region, and the vendor's own price at purchase time. ## First call 1. GET /healthz — if book.stale or payments.ok is false, stop. Do not pay. 2. GET /v1/example — shape only, example: true. 3. GET /v1/cheapest — unpaid 402. Pay from that accepts list only (Base or Solana). 4. Resend with PAYMENT-SIGNATURE. 200 or paid 404 is the answer. One paid route. 0.005 USDC. No key. ## Provider fleet The current registry covers 16 companies in 17 feeds: 11 inference and 6 GPU. | provider | work | source | |---|---|---| | openai | inference | https://platform.openai.com/docs/pricing | | anthropic | inference | https://docs.claude.com/en/docs/about-claude/pricing | | google | inference | https://ai.google.dev/gemini-api/docs/pricing | | groq | inference | https://console.groq.com/docs/models | | together | inference | https://www.together.ai/pricing | | deepseek | inference | https://api-docs.deepseek.com/quick_start/pricing | | mistral | inference | https://mistral.ai/pricing/api/ | | fireworks | inference | https://docs.fireworks.ai/serverless/pricing | | xai | inference | https://docs.x.ai/developers/models | | runpod | gpu | https://www.runpod.io/pricing | | lambda | gpu | https://lambda.ai/service/gpu-cloud | | fireworks | gpu | https://fireworks.ai/pricing | | crusoe | gpu | https://www.crusoe.ai/cloud/pricing | | minimax | inference | https://platform.minimax.io/docs/guides/pricing-paygo | | cerebras | inference | https://api.cerebras.ai/public/v1/models | | nebius | gpu | https://nebius.com/prices | | hyperstack | gpu | https://www.hyperstack.cloud/gpu-pricing | ## Free vs paid These endpoints are free: GET /v1/example GET /openapi.json GET /.well-known/x402 GET /llms.txt GET /skill.md GET /healthz GET /privacy GET /privacy.txt Only GET /v1/cheapest is paid: 0.005 USDC per call. It uses x402 v2 on Base or Solana with the PayAI facilitator. ## When to use this Use it when you need to choose where to run a workload and the choice turns on price: picking a model for a known token budget, or picking a GPU cloud for a known number of GPU-hours. Do not use it for: - Pricing State — https://pricing-state-api.replit.app/ — What is this vendor's seat / list ladder? - Status State — https://status-state-api.replit.app/ — Is the vendor up? - Changelog State — https://changelog-state-api.replit.app/ — Did the SDK ship a break? ## Learn the shape for free curl https://compute-state-api.replit.app/v1/example That returns a fixed sample of the paid body, flagged example: true. ## Ask a real question Embeddings, priced on input alone: curl "https://compute-state-api.replit.app/v1/cheapest?work=inference&model_tier=embed&est_input_tokens=100000" An overnight batch job where most of the prompt is a cached prefix: curl "https://compute-state-api.replit.app/v1/cheapest?work=inference&batch=true&est_input_tokens=200000&cached_input_tokens=180000&est_output_tokens=5000" Both return 402 the first time. The challenge body lists both chains under accepts; the same JSON is base64-encoded in the PAYMENT-REQUIRED header. Sign the payment payload for whichever chain you hold USDC on and resend the request with it base64-encoded in PAYMENT-SIGNATURE. The 200 carries the settlement receipt in PAYMENT-RESPONSE. ## Paying Price is 0.005 USDC per settled call, on Base (eip155:8453) or Solana, x402 v2 scheme exact, settled by the PayAI facilitator. Build your payment from the challenge's own accepts list — do not hardcode the token's name or a blockhash, both change over time. You need a wallet that can complete that challenge: EIP-3009 transferWithAuthorization on Base USDC, or a Solana USDC transfer built against the challenge's extra.feePayer and recentBlockhash. Use any x402 v2 client pointed at this challenge. PayAI is the facilitator. There is no API key for this product. If no payout wallet is configured on this deployment, GET /v1/cheapest answers 503, not a 402. 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. ## Reading the answer - recommendation.est_cost_usd is the total for the job you described, not a rate. - recommendation.components shows how that total was assembled: each entry has the published rate (rate_usd), its unit, its contribution (usd), and a tag. - A tag of unknown never appears on the top recommendation. If a SKU had an unreadable component it is in disqualified with the component named. - as_of is the oldest input timestamp. Anything older than 900s cannot be used in an answer. - disqualified is not noise. It is the list of things that matched your question and could not be ranked, with the reason on each. ## Failure modes worth handling - 400 — your query was invalid. The message says what. Unpaid. Fix and retry. Sent only once a signed payment verifies; an unsigned call gets the 402. - 402 — pay, or your signature was rejected at verify or settle. Unpaid. Retry with a fresh payload — unless the body says settlement is pending, in which case resend the identical PAYMENT-SIGNATURE. - 503 — the book is stale or empty. Unpaid. Honour Retry-After. - 404 — wrong path (unpaid), or a verified call whose book had no match once ranked, then settled (paid); disqualified explains what came close. - 500 — rare: the process faulted before settlement. Operator fault. You were not charged. Not something a retry with different parameters fixes. No body, no settle; no settle, no charge. 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. The book we rank is the one that passed the pre-pay freshness check, even if the clock crossed 900s during the wallet step. ## Constraints to respect - work is required and must be exactly one of inference or gpu. Sending inference parameters with work=gpu (or the reverse) is a 400 — this API answers one question per call and will not optimise a mixed stack. - availability=spot is a 400. Spot is not ingested in v1. - region is accepted as a hint and does not filter in v1. - batch and cached_input_tokens are inference-only; sending either with work=gpu is a 400, as is cached_input_tokens greater than est_input_tokens. - batch=true and cached_input_tokens > 0 rank on rates the vendor itself publishes. A SKU without the rate your query needs is disqualified (no_official_batch_rate, no_official_cached_input_rate), never silently priced at its interactive rate. The exception is a vendor that publishes that it charges the standard input rate for cached tokens (Cerebras), where the cached share is priced at that published input rate and named cached_input_at_input_rate. - model_tier=embed is input-only: est_output_tokens is ignored rather than rejected, and only official embedding SKUs compete. - DeepSeek is billed by the clock — peak (01:00-04:00 and 06:00-10:00 UTC, Monday to Friday) and off-peak are separate published rates, so the same query can cost different amounts at different hours. recommendation.why names the window used. - gpu_model matches a GPU family, not a product string: h100 covers H100 SXM and H100 PCIe as separate SKUs; h200 never matches an H100. - model_tier buckets come from published output price: cheap is <= $1.00/1M, frontier is >= $10.00/1M. Prices between the two are in neither bucket and will not be returned under a tier filter.