Advanced Chat Completion Generation API

What it does

Chat Completions lets you create a model-driven chat response from a message list. Send a required model and messages, then tune generation with options like temperature, top_p, max_tokens, stream, n, stop, seed, and penalties for presence or frequency.

Use it when you need a single API surface for conversational AI, structured output, tool calling, or search-augmented responses. The request schema also supports tools, tool_choice, parallel_tool_calls, web_search, plugins, response_format, guardrails, and image_config, so you can adapt the same endpoint to text-only chats, JSON-shaped responses, or multimodal workflows that include images in messages.

The response includes an id, model, created timestamp, and a choices array. Each choice returns a message or streaming delta payload with fields such as role, content, reasoning, tool_calls, annotations, and tool_call_id, plus finish_reason for completion status.

You also get a usage object with token and cost breakdowns, including prompt_tokens, completion_tokens, reasoning_tokens, cached_tokens, and cost fields for input, output, reasoning, web search, audio input, cached input, and request total. That makes Chat Completions suitable for production systems that need both model output and accounting data.