Chat Completions

{
  "id": "chatcmpl_...",
  "object": "chat.completion",
  "created": 1735689600,
  "model": "studio/qwen3-0.6b",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "Hi." },
      "finish_reason": "stop"
    }
  ],
  "usage": { "prompt_tokens": 18, "completion_tokens": 3, "total_tokens": 21 },
  "llm_studio": {
    "request_id": "req_...",
    "credits_consumed": 21
  }
}

llm_studio is additive metadata outside the compatibility-critical fields -- existing OpenAI SDKs ignore it safely. credits_consumed is 0 for BYOK models, since BYOK usage is billed by the provider directly and never draws down hosted credits.

Pass tools the same way as OpenAI's API. Only models with supports_tools: true (see Models) accept it -- sending tools to a model that doesn't support them returns 400 invalid_request_error before any request reaches the model.

{
  "model": "openai/gpt-5.1",
  "messages": [{"role": "user", "content": "What's the weather in Paris?"}],
  "tools": [
    {
      "type": "function",
      "function": {
        "name": "get_weather",
        "parameters": { "type": "object", "properties": { "city": { "type": "string" } } }
      }
    }
  ]
}

Messages also support a tool role and tool_calls / tool_call_id, for feeding a tool's result back into the conversation -- the same shape as OpenAI's API. Image content parts ({"type": "image_url", "image_url": {"url": "..."}}) work the same way for models with supports_vision: true.

For non-streaming requests, pass an Idempotency-Key header to safely retry a request -- a replayed request with the same key returns the original response instead of generating again. Streaming requests don't support idempotency keys today.

Streaming responses are documented separately -- see Streaming. Error shapes are documented under Errors. For OpenAI's newer Responses API shape, see Responses API.