Reasoning parameters

The reasoning object

Field	Type	Default	Notes
`enabled`	boolean		Turn reasoning on.
`effort`	string	`low`	One of `low`, `medium`, `high`. How hard the model thinks.
`summary`	string	`auto`	One of `auto`, `concise`, `detailed`. Granularity of the returned reasoning summary.
`max_tokens`	integer		Cap on reasoning tokens.
`exclude`	boolean		When set, suppress reasoning details from the output.

{
  "model": "z-ai/glm-5.2",
  "messages": [ { "role": "user", "content": "Plan a three-step migration." } ],
  "reasoning": { "enabled": true, "effort": "high", "summary": "detailed" }
}

Reasoning can also be requested with a model-ID feature suffix; the request object gives per-call control over effort and summary.

Output

The response message carries reasoning_details and usage reports reasoning_tokens, billed like other tokens. Set exclude: true to reason without returning the details.

reasoning_details

Field	Type	Notes
`type`	string	The kind of reasoning detail.
`summary`	string	The summary text for this segment.
`id`	string	Identifier for the segment.
`format`	string	The format of the detail.
`index`	integer	Ordering index.

{
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "...",
        "reasoning_details": [
          { "type": "...", "summary": "First I considered...", "id": "...", "format": "...", "index": 0 }
        ]
      },
      "finish_reason": "stop"
    }
  ],
  "usage": { "prompt_tokens": 30, "completion_tokens": 80, "reasoning_tokens": 140, "total_tokens": 250 }
}

Streaming

In a stream, reasoning_details arrives on deltas with a non-empty summary, in the same shape. See Delta format.

Effort shorthand

Top-level reasoning_effort (minimal, low, medium, high) is accepted for OpenAI-style clients; minimal normalizes to low, and reasoning.effort wins when both are sent.

Basics

Features

Server-Side Tools

Multimodal

Best Practices

SDKs

API Reference

Reasoning parameters

The reasoning object

Output

reasoning_details

Streaming

Effort shorthand

​The reasoning object

​Output

​reasoning_details

​Streaming

​Effort shorthand

​Related

The reasoning object

Output

reasoning_details

Streaming

Effort shorthand

Related