Thinking Mode

DeepSeek models support a "thinking" mode where the model reasons step-by-step before answering.

Enable Thinking

Add the thinking parameter to your request to enable step-by-step reasoning.

{
  "model": "deepseek-v4-pro",
  "messages": [
    {"role": "user", "content": "Solve: x² + 5x + 6 = 0"}
  ],
  "thinking": {"type": "enabled"},
  "reasoning_effort": "high"
}

Reasoning Effort Levels

• low — minimal reasoning, fastest response • medium — balanced reasoning • high — thorough reasoning (default when thinking is enabled) • max — deepest reasoning for the most complex tasks

Reading Thinking Content

When thinking is enabled, the response includes a reasoning_content field in addition to the regular content field. The reasoning_content shows the model's chain-of-thought process.