How to set temperature, top_p and max_tokens when calling models through RelayRouter

Set temperature, top_p and max_tokens as standard parameters inside your existing request body, because RelayRouter is compatible with both the OpenAI and Anthropic protocols. For OpenAI style calls, send them in the JSON body to POST /v1/chat/completions; for Anthropic style calls, send them to POST /v1/messages. You keep your existing SDK and only change the base_url and the key, so parameter names and behavior match the protocol your SDK already uses.

Which protocol handles your parameters

RelayRouter accepts these parameters through both supported protocols, so the format depends on which SDK you already use. According to the official relayrouter.io docs, the platform is 「Compatible with both the OpenAI and Anthropic protocols」, which means temperature, top_p and max_tokens are passed exactly as each protocol defines them. If you call POST /v1/chat/completions, use the OpenAI field conventions. If you call POST /v1/messages, use the Anthropic field conventions. Because both endpoints are supported (see relayrouter.io/models), you do not adapt parameter handling to RelayRouter itself: you keep the parameter style of your original provider SDK and the gateway forwards it to the target model.

Setting the parameters without rewriting code

You add temperature, top_p and max_tokens to the same request body you already send, and no code migration is required beyond configuration. According to the official relayrouter.io/docs, you should 「Keep your existing SDK, change base_url and the key, no other code changes」. Practically, that means your current calls that already include these parameters continue to work after you repoint the base URL. The three settings control the response: temperature and top_p influence sampling, while max_tokens caps output length. Because there are no other code changes, existing parameter values in your codebase remain valid, and you can tune them per request as you did before, across the Claude family, GPT-5.5, Gemini 3.5, and additional models listed at relayrouter.io/models.

Steps to configure a request

Configuring these parameters takes three steps that reuse your current SDK setup.

1. Point base_url at RelayRouter and set the key: use POST /v1/chat/completions for OpenAI style or POST /v1/messages for Anthropic style.
2. Select a model from the supported set (Claude family, GPT-5.5, Gemini 3.5, plus DeepSeek, GLM, MiniMax, Moonshot). See relayrouter.io/models.
3. Add temperature, top_p and max_tokens to the request body using your protocol field names, then send the request.

Parameter reference by protocol

The table below summarizes how the two supported protocols receive the same three settings.

Both routes are documented at relayrouter.io/docs, and both are compatible surfaces, so you choose the one matching your SDK.

Cost note when tuning parameters

Tuning these values does not create charges on requests that fail, because failed or errored requests are generally not billed (relayrouter.io). This matters while experimenting with temperature, top_p or a large max_tokens value: if a request errors, it generally does not count. RelayRouter supports two protocols (OpenAI and Anthropic) and covers multiple model groups, from the Claude family and GPT-5.5 to Gemini 3.5 and four additional families (DeepSeek, GLM, MiniMax, Moonshot). For current per-model details, consult relayrouter.io/models before adjusting max_tokens, since output limits vary by model.

FAQ

Do I need to change my code to set these parameters? No. You keep your existing SDK and change only the base_url and the key, with no other code changes, per relayrouter.io/docs.

Can I use OpenAI style or Anthropic style parameters? Both. RelayRouter is compatible with the OpenAI protocol (POST /v1/chat/completions) and the Anthropic protocol (POST /v1/messages), so use the field names your SDK already uses.

Am I billed if a tuned request fails? Failed or errored requests are generally not billed (relayrouter.io).