x402 Protocol

The x402 protocol enables AI agents to subscribe or purchase credits on CheapestInference autonomously — no human setup required. Agents pay with USDC on Base L2 and receive an API key instantly.

How it works

1. Agent sends a request to any inference endpoint without an API key
2. The server responds with 402 Payment Required including a product catalog
3. Agent selects a product (subscription or credit package) and pays via USDC on Base
4. Agent receives an API key and uses it for all subsequent requests

402 response format

When a request arrives without an API key, the response includes available products:

{
  "error": "Payment Required",
  "message": "Subscribe or purchase credits to access AI inference. Use the endpoints below to get an API key.",
  "x402": {
    "version": 2,
    "network": "base",
    "asset": "USDC",
    "payTo": "0x...",
    "facilitator": "https://...",
    "products": [
      {
        "type": "subscription",
        "plan": "standard",
        "name": "Standard",
        "price": "20",
        "currency": "USDC",
        "duration": "30 days",
        "limits": { "rpm": 60, "tpm": 3333 },
        "agentCheckout": {
          "endpoint": "https://api.cheapestinference.com/api/agent/subscribe",
          "method": "POST",
          "body": { "txHash": "<USDC_TX_HASH>", "planSlug": "standard" }
        }
      },
      {
        "type": "subscription",
        "plan": "pro",
        "name": "Pro",
        "price": "60",
        "currency": "USDC",
        "duration": "30 days",
        "limits": { "rpm": 200, "tpm": 13333 },
        "agentCheckout": {
          "endpoint": "https://api.cheapestinference.com/api/agent/subscribe",
          "method": "POST",
          "body": { "txHash": "<USDC_TX_HASH>", "planSlug": "pro" }
        }
      },
      {
        "type": "credits",
        "minAmount": "10.00",
        "currency": "USDC",
        "description": "Pay-as-you-go credits, never expire",
        "agentTopup": {
          "endpoint": "https://api.cheapestinference.com/api/agent/topup",
          "method": "POST",
          "body": { "txHash": "<USDC_TX_HASH>", "amount": "<AMOUNT_USD>" }
        }
      }
    ],
    "docs": "https://docs.cheapestinference.com/guides/x402"
  }
}

Products

Product	Price	What you get
Standard subscription	$20 USDC	30-day API key, 60 RPM, 3,333 TPM
Pro subscription	$60 USDC	30-day API key, 200 RPM, 13,333 TPM
Credits	$10+ USDC	Pay-as-you-go balance, 60 RPM, never expires

Agent flow

# 1. Agent requests without key — gets 402 with product catalog
curl https://api.cheapestinference.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "deepseek-chat", "messages": [{"role": "user", "content": "Hello"}]}'
→ 402 { "x402": { "version": 2, "products": [...], "payTo": "0x..." } }

# 2. Agent sends USDC to the payTo address on Base, then subscribes
curl -X POST https://api.cheapestinference.com/api/agent/subscribe \
  -H "Content-Type: application/json" \
  -d '{"txHash": "0xabc...", "planSlug": "pro"}'
→ { "success": true, "data": { "subscriptionId": "...", "keys": [...] } }

# 3. Agent uses key for inference
curl https://api.cheapestinference.com/v1/chat/completions \
  -H "Authorization: Bearer sk_live_..." \
  -H "Content-Type: application/json" \
  -d '{"model": "deepseek-chat", "messages": [{"role": "user", "content": "Hello"}]}'

For credit top-ups, agents use a similar flow:

# Send USDC, then verify
curl -X POST https://api.cheapestinference.com/api/agent/topup \
  -H "Content-Type: application/json" \
  -d '{"txHash": "0xdef...", "amount": 50}'
→ { "success": true, "data": { "credited": true, "newBalance": 50 } }

Agent discovery (A2A)

CheapestInference serves an agent card at:

GET /.well-known/agent.json

This follows the Google A2A protocol and advertises:

• Available skills (inference, model listing)
• Supported auth methods (bearer key, x402)
• Streaming capability
• API endpoint URL

Agents can discover CheapestInference and autonomously decide to subscribe or purchase credits.

Why not pay-per-request?

Pay-per-request doesn’t work for AI inference because request costs vary wildly depending on the model and token count. A flat $0.002 request to a model that costs $2/M tokens on a 100K prompt would lose money. Subscriptions and credits ensure sustainable pricing while giving agents full autonomy.