Core features

These guides cover the request features shared by every model: how to stream output, let models call your functions, get typed JSON back, control reasoning, and handle errors. Examples use the native Responses API; the same features are available through the OpenAI- and Anthropic-compatible endpoints described in SDK compatibility.

Guides

• Streaming — Receive output as it is generated
• Tool Use — Let models call functions in your application
• Structured Output — Get JSON that matches a schema
• Reasoning Effort — Trade latency and cost for deeper reasoning
• Error Handling — Handle errors with retries and fallbacks