Chat Completions
POST /v1/chat/completions
Required
model: canonical catalog ID; messages: nonempty array of system, developer, user, assistant or tool messages with text/null content.
Optional
stream, max_tokens or max_completion_tokens, temperature, top_p, stop, tools, reasoning_effort, thinking and stream_options.include_usage. Only n: 1 is supported.
Reasoning
minimal, low, medium, high, xhigh, max and ultra are policy levels. Only select levels advertised for the model. Thinking budget must leave output headroom; conflicting selectors return 400.