Send the user turn and your route options; get back one choice and a probability for each option. Your code decides what to do with it: send easy prompts to a cheaper model, hard ones to a stronger one.
The routes are yours: name them after your own models or queues.
Output tokens are not billed, so a decision costs only the input you send.
Cost example
Illustrative cost for 1,000 decisions
List-price illustration for accounts created on or after 12 November 2026: 1,000 text decisions, one question per request and 1,000 billed input tokens each, cost 0.028 EUR on s1-fast (Peer-to-Peer) or 0.030 EUR (EU); s1-pro costs 0.085 EUR or 0.090 EUR. Output tokens are free; applicable taxes are excluded. Input tokens include the state, instructions and answer options. Actual costs depend on the input and tier. Full pricing.
FAQ
Frequently asked question
Which model should I use for LLM routing?
Choose s1-fast for one typed routing question per request. Use s1-pro when a routing decision needs up to three typed text questions in one request. Set your own routes and validate them on representative prompts before automating.