RouterBench

•  the gateway  •

One API.
Every model.

Point your app at RouterBench and reach 130+ models across 25+ providers through a single OpenAI-compatible endpoint — with smart routing, failover, caching, guardrails and per-request cost tracking built in.

app.ts
const client = new OpenAI({
  baseURL: "https://api.routerbench.com/v1",
  apiKey: process.env.ROUTERBENCH_API_KEY,
});

// model: "routerbench/auto" → the router decides
// or pin any model: "anthropic/claude-opus"

•  drop-in  •

Change one string. Keep everything else.

  • compatible Your OpenAI SDKs, agents and tools keep working
  • flexible BYOK, RouterBench credits, or hybrid — per project
  • transparent Per-request cost in USD on every response
  • observable OpenTelemetry tracing out of the box

•  endpoints  •

The full surface — not workarounds through chat.

/v1/chatchat completions — streaming, tools, vision
/v1/embeddingsembeddings
/v1/modelsthe live model catalog
/v1/messagesAnthropic-native messages
/v1/audioTTS · STT · translation
/v1/realtimerealtime voice over WebSocket
/v1/imagesimage generation
/v1/musicmusic generation
/v1/videosvideo generation
/v1/responsesresponses API

•  built in  •

Production plumbing, included.

01 — reliability

Failover & retries

Automatic provider failover, retries with backoff and circuit breakers — no vigil required.

02 — cost

Response & prompt caching

Serve repeats instantly at zero inference cost; prompt caching cuts input costs on long contexts.

04 — output

Structured outputs + healing

JSON schema outputs, with automatic repair of malformed responses — 5 healing strategies.

05 — tools

Web search, vision, tools

Built-in web search, native tool calling and vision across supported models.

06 — control

Budgets & rate limits

Per-key budgets, model allow-lists and rate limits — scope every key's blast radius.

Integrate once. Reach everything.

Get an API key and make your first routed request in minutes.