Docs / Layer 1 Oracle Data / Compute & Inference

Compute & Inference Pricing

Signed GPU compute and LLM inference pricing. Cloud GPU spot rates and model inference costs — cryptographically attested on every query.

Overview

Two data feeds: GPU compute pricing across major cloud providers, and LLM inference pricing across frontier model APIs. Both signed with Ed25519 on every paid query.

Useful for agents that need to make cost-aware decisions — which cloud to provision on, which model to use for a given task, or whether inference costs have shifted since last check.

GPU Compute Endpoints

EndpointDescriptionPricePreview
/oracle/compute/allAll GPU instances — H100 SXM/PCIe, A100, A10G, L4, T4 across AWS, GCP, Azure, CoreWeave, Lambda$0.05/preview ↗
/oracle/compute/best/h100_sxmCheapest H100 SXM instance available right now$0.02/preview ↗
/oracle/compute/best/a100Cheapest A100 instance available right now$0.02/preview ↗
/oracle/compute/compareSide-by-side price comparison across providers for a given GPU type$0.05/preview ↗

LLM Inference Endpoints

EndpointDescriptionPricePreview
/oracle/inference/allAll tracked models — GPT-4o, Claude Sonnet/Opus, Gemini Pro/Flash, Llama 3, Mistral. Input + output cost per million tokens.$0.02/preview ↗
/oracle/inference/best/codingBest price-performance model for coding tasks right now$0.02/preview ↗
/oracle/inference/best/generalBest price-performance model for general tasks$0.02/preview ↗
/oracle/inference/compareModel pricing comparison — cost per million tokens, context window, latency tier$0.02/preview ↗

Response Format

GPU compute example:

json
{
  "gpu": "h100_sxm",
  "best": {
    "provider": "lambda",
    "price_per_hour": 2.49,
    "currency": "USD",
    "region": "us-east-1",
    "availability": "available"
  },
  "all": [
    {"provider": "lambda", "price_per_hour": 2.49, "region": "us-east-1"},
    {"provider": "coreweave", "price_per_hour": 2.79, "region": "us-east-1"},
    {"provider": "aws", "price_per_hour": 3.84, "region": "us-east-1"}
  ],
  "fetched_at": 1781100000,
  "signed": true
}

Inference pricing example:

json
{
  "models": [
    {
      "provider": "anthropic",
      "model": "claude-sonnet-4",
      "input_per_mtok": 3.00,
      "output_per_mtok": 15.00,
      "context_window": 200000
    },
    {
      "provider": "openai",
      "model": "gpt-4o",
      "input_per_mtok": 2.50,
      "output_per_mtok": 10.00,
      "context_window": 128000
    }
  ],
  "fetched_at": 1781100000,
  "signed": true
}

Payment

All compute and inference endpoints are x402 only (USDC on Base). L402 Lightning is not available for these endpoints. Append /preview to any endpoint for unsigned sample data — no payment required.

Use case: An agent deciding which model to use for a task can call /oracle/inference/all ($0.02) once per session to get current pricing, then make cost-aware routing decisions for the session duration without re-querying.