Thinking Variant Documentation

The :thinking variant enables extended reasoning capabilities for complex problem-solving tasks.

Usage

Append :thinking to any model ID:

Example Request:

{
  "model": "deepseek/deepseek-r1:thinking"
}

💡 Quick Tip

Simply add :thinking as a suffix to your model identifier to enable extended reasoning mode. For example:

deepseek/deepseek-r1 → deepseek/deepseek-r1:thinking
anthropic/claude-3-opus → anthropic/claude-3-opus:thinking

Details

Thinking variants provide access to models with extended reasoning capabilities, allowing for more thorough analysis and step-by-step problem solving. This is particularly useful for complex tasks that benefit from chain-of-thought reasoning.

                When to Use Thinking Variants
                Complex Problem Solving: Tasks that require multi-step reasoning and analysis
Mathematical Problems: Calculations and proofs that need step-by-step breakdown
Logical Reasoning: Scenarios requiring thorough chain-of-thought processes
Code Generation: Complex algorithms that benefit from detailed planning
Analysis Tasks: Breaking down complex information into understandable parts

            

How It Works

When you use the :thinking variant, the model will:

Engage in extended internal reasoning before producing a response
Break down complex problems into smaller, manageable steps
Use chain-of-thought processes to arrive at solutions
Provide more thorough analysis and consideration of multiple approaches

📖 Related Documentation

See also: Reasoning Tokens for more information about how reasoning tokens are handled and billed.

Implementation Example

Here's a complete example of how to use the thinking variant in an API request:

Complete API Request:

{
  "model": "deepseek/deepseek-r1:thinking",
  "messages": [
    {
      "role": "user",
      "content": "Solve this complex problem step by step: [your problem]"
    }
  ],
  "temperature": 0.7,
  "max_tokens": 4096
}

Performance Considerations

When using thinking variants, keep in mind:

Response Time: Extended reasoning may increase response latency
Token Usage: Reasoning tokens are typically counted separately (see Reasoning Tokens documentation)
Cost: Extended reasoning may result in higher token usage and associated costs
Use Cases: Best suited for complex problems that truly benefit from extended reasoning