Reasoning

POST

/v1/chat/completions

Reasoning-optimised models use extended internal deliberation before responding, trading latency for higher accuracy on complex tasks.

Request body

Reasoning models share the standard chat completions request shape. Some models also accept the optional fields below — treat them as best-effort and confirm support per model.

Field	Type	Required	Notes
`model`	string	Yes	A reasoning-capable model id from `/v1/models`.
`messages`	array	Yes	`user` and `assistant` turns. System prompts may be limited on some reasoning models.
`max_completion_tokens`	integer	No	Where supported, controls the total token budget (reasoning tokens + output tokens). Some models use `max_tokens` instead.
`reasoning_effort`	string	No	`low`, `medium`, or `high` where supported. Higher effort = more thorough reasoning at greater cost. Unsupported models ignore or reject it.

Example request

{
"model": "myt/qwen3.6-35b-a3b",
"messages": [
  {
    "role": "user",
    "content": "What is the fastest sorting algorithm for nearly-sorted arrays and why?"
  }
],
"max_completion_tokens": 2048
}

Response

{
"id": "chatcmpl-reason123",
"object": "chat.completion",
"created": 1749366600,
"model": "myt/qwen3.6-35b-a3b",
"choices": [
  {
    "index": 0,
    "finish_reason": "stop",
    "message": {
      "role": "assistant",
      "content": "For nearly-sorted arrays, Insertion Sort is the fastest in practice with O(n + d) time complexity, where d is the number of inversions..."
    }
  }
],
"usage": {
  "prompt_tokens": 22,
  "completion_tokens": 312,
  "total_tokens": 334
}
}

When a model reports its internal reasoning usage, it appears under usage.completion_tokens_details.reasoning_tokens. Those tokens count against your budget but are not returned in the message content. Not every model reports this field.

Errors

Status	When
`400`	Unsupported parameter for the chosen model (e.g. `temperature`, `stream`, `reasoning_effort`)
`401`	API key missing or invalid
`403`	Model not enabled for your project
`429`	Rate limit or budget exceeded

Code examples

curl -sS --request "POST" \
  --url "$MYT_BASE_URL/chat/completions" \
  --header "Authorization: Bearer $MYT_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{"model":"myt/qwen3.6-35b-a3b","messages":[{"role":"user","content":"What is the fastest sorting algorithm for nearly-sorted arrays and why?"}],"max_completion_tokens":2048}'

Full parameter reference: Chat completions.