Example: AI API Reseller

Resell access to the Anthropic Claude API. Clients pay per-request via x402 — pricing adjusts automatically based on which model they request.

Use case

You hold an Anthropic API key and want to monetize it. Instead of managing API keys, subscriptions, or billing dashboards, you put tollbooth in front of the Anthropic API and charge per-request in USDC. Cheaper models cost less, expensive models cost more.

There are two ways to price this: flat per-request or token-based. You can pick whichever fits your business.

For production operations (model table updates, fine-tune mapping, safe rollouts, and metrics), see LLM Pricing Operations.

Option A: Flat per-request pricing

Charge a fixed amount per API call, regardless of how many tokens the request uses.

gateway:
  port: 3000
  discovery: true

wallets:
  base: "0xYourWalletAddress"

accepts:
  - asset: USDC
    network: base

upstreams:
  anthropic:
    url: "https://api.anthropic.com"
    headers:
      x-api-key: "${ANTHROPIC_API_KEY}"
      anthropic-version: "2023-06-01"
    timeout: 120

routes:
  "POST /v1/messages":
    upstream: anthropic
    path: "/v1/messages"
    match:
      - where: { body.model: "claude-haiku-*" }
        price: "$0.005"
      - where: { body.model: "claude-sonnet-*" }
        price: "$0.015"
      - where: { body.model: "claude-opus-*" }
        price: "$0.075"
    fallback: "$0.015"

What’s going on

Single upstream pointing at api.anthropic.com with your API key injected via env var.
One route (POST /v1/messages) mirrors the Anthropic Messages endpoint.
Match rules inspect body.model using glob patterns to set per-model pricing — Haiku is cheap, Opus is premium.
Fallback catches any model that doesn’t match a rule (e.g. new models Anthropic releases).

Option B: Token-based pricing

Price per-request based on which model is used. Set type: token-based and tollbooth reads the model field from the request body automatically — no match rules needed.

gateway:
  port: 3000
  discovery: true

wallets:
  base: "0xYourWalletAddress"

accepts:
  - asset: USDC
    network: base

upstreams:
  anthropic:
    url: "https://api.anthropic.com"
    headers:
      x-api-key: "${ANTHROPIC_API_KEY}"
      anthropic-version: "2023-06-01"
    timeout: 120

routes:
  "POST /v1/messages":
    upstream: anthropic
    path: "/v1/messages"
    type: token-based
    models:
      claude-haiku-4-5-20251001: "$0.001"
      claude-sonnet-4-5-20250929: "$0.005"
      claude-opus-4-6: "$0.02"
    fallback: "$0.005"

What’s going on

type: token-based tells tollbooth to extract the model from the request body and look up a per-request price. (openai-compatible is also accepted as an alias.)
models map sets a flat per-request price for each model. The price is charged once per API call, not per token.
Fallback applies a default per-request price for any model not explicitly listed.
No match rules required — the gateway handles model detection automatically.

Which should you pick?

	Flat per-request (match rules)	Token-based (model lookup)
Simplicity	You write explicit match rules for each model family	Gateway auto-detects model and looks up price — no match rules
Flexibility	Glob patterns match model families (e.g. `claude-sonnet-*`)	Exact model names from a built-in table + your overrides
Maintenance	Update match rules when new models launch	Built-in table covers common models; add overrides as needed
Pricing	Both charge a flat per-request price	Both charge a flat per-request price

Rule of thumb: use match rules when you want to price by model family with glob patterns. Use token-based when you want automatic per-model pricing without writing match rules.

Run it

export ANTHROPIC_API_KEY="sk-ant-..."
npx tollbooth start

Expected flow

Client                        Tollbooth                     Anthropic
  │                              │                              │
  │  POST /v1/messages           │                              │
  │  { model: "claude-sonnet-…"} │                              │
  │─────────────────────────────>│                              │
  │                              │  match body.model →          │
  │                              │  price: $0.015               │
  │  402 + payment instructions  │                              │
  │<─────────────────────────────│                              │
  │                              │                              │
  │  (sign $0.015 USDC payment)  │                              │
  │                              │                              │
  │  POST /v1/messages           │                              │
  │  + payment-signature header   │                              │
  │─────────────────────────────>│                              │
  │                              │  verify + settle payment     │
  │                              │                              │
  │                              │  POST /v1/messages           │
  │                              │──────────────────────────────>│
  │                              │                              │
  │                              │  { content: "..." }          │
  │                              │<──────────────────────────────│
  │  200 + Claude response       │                              │
  │<─────────────────────────────│                              │

Try it with curl

First request — get the 402:

curl -s http://localhost:3000/v1/messages \
  -X POST \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-sonnet-4-20250514","max_tokens":256,"messages":[{"role":"user","content":"Hello"}]}'

The response includes a 402 Payment Required status with payment instructions in the headers. An x402-compatible client (or wallet SDK) reads these instructions, signs the USDC payment, and resends the request with the payment proof attached.

Next: Video Streaming Paywall →