Now live

Pay top-tier prices
only for hard requests

Not every request needs your most expensive model. A quick question and a hard, multi-step task cost the same per token if they go to the same model. Auto routing looks at each new conversation and sends it to the model you chose for that kind of request.

Auto routing: a first message is scored against three routes and sent to Haiku 5.5, Sonnet 5.5 or Opus 5.5

Routes from the policy we use for our own coding agents.

Up to 43%

lower cost on coding tasks

Up to 30%

lower latency

Internal benchmarks against sending everything to the strongest model. Your results may vary.

How it works

Describe your requests in plain language

Write each type in plain language and pick a model for it. Or start from a template.

A decision model picks the route

A decision model reads the first message and chooses. If it isn’t sure, your fallback model takes it.

The choice sticks for the session

The whole conversation stays on one model, so prompt caching keeps working.

Test it before you ship

Try real messages first. A test costs a fraction of a cent.

Use it from Claude Code

export ANTHROPIC_BASE_URL="https://router.requesty.ai"
export ANTHROPIC_AUTH_TOKEN="<your Requesty key>"
export ANTHROPIC_MODEL="policy/<your-policy>"
claude

Anywhere else, set model to policy/<your-policy>. It works with the Chat Completions, Messages and Responses APIs, including streaming and tools.

Create an auto policy

Read the docs

Questions? Contact support@requesty.ai or read the docs.

Unsubscribe

Pay top-tier pricesonly for hard requests

How it works

Use it from Claude Code

Pay top-tier prices
only for hard requests