# Reasoning
Grok 4 and other reasoning models provide advanced reasoning capabilities for complex problem-solving tasks.
## Overview
Reasoning models like Grok 4 are designed to:
- Break down complex problems into steps
- Show their thought process
- Provide more accurate answers for difficult tasks
- Handle multi-step reasoning and analysis
## Grok 4 Reasoning
Grok 4 is a reasoning model with **no non-reasoning mode available**. It always applies extended reasoning to generate responses.
### Important Notes
- `reasoning_effort` parameter is **not supported** by Grok 4. Providing it will result in an error.
- The `presencePenalty`, `frequencyPenalty`, and `stop` parameters are **not supported** by reasoning models.
## Using Reasoning Models
```python
from openai import OpenAI
client = OpenAI(
api_key=os.environ.get("XAI_API_KEY"),
base_url="https://api.x.ai/v1"
)
response = client.chat.completions.create(
model="grok-4",
messages=[
{
"role": "user",
"content": "Solve this step by step: If a train travels 120 miles in 2 hours, then slows down and travels 80 miles in 2 more hours, what is the average speed for the entire journey?"
}
]
)
print(response.choices[0].message.content)
```
## Reasoning vs Non-Reasoning Models
| Feature | Reasoning Models | Non-Reasoning Models |
|---------|------------------|---------------------|
| Extended thinking | Yes | No |
| Complex problem solving | Excellent | Good |
| Response speed | Slower | Faster |
| Token usage | Higher | Lower |
| Best for | Analysis, math, coding | Quick responses, simple tasks |
## Model Variants
### Grok 4
- Full reasoning capability
- No non-reasoning mode
- 2M context length
- Best for complex tasks
### Grok 4.1 Fast Reasoning
- Reasoning enabled
- Optimized for speed
- 131K context length
- Good for agentic tool calling
### Grok 4.1 Fast Non-Reasoning
- No extended reasoning
- Fastest responses
- 131K context length
- Best for simple tasks
### Grok 4 Fast Reasoning
- Balanced reasoning
- Cost-efficient
- Good for moderate complexity
### Grok 4 Fast Non-Reasoning
- No extended reasoning
- Very fast
- Best for high-throughput simple tasks
## When to Use Reasoning Models
**Use reasoning models for:**
- Mathematical problems
- Logic puzzles
- Code debugging and analysis
- Multi-step planning
- Complex document analysis
- Scientific reasoning
**Use non-reasoning models for:**
- Simple Q&A
- Text summarization
- Quick translations
- High-volume simple tasks
- Cost-sensitive applications
## Reasoning Tokens
Reasoning models may generate "reasoning tokens" that represent the model's thought process. These tokens contribute to the total token count and cost.
```json
{
"usage": {
"prompt_tokens": 50,
"completion_tokens": 200,
"reasoning_tokens": 150,
"total_tokens": 400
}
}
```
## Best Practices
1. **Be specific**: Clearly state what you want the model to reason about
2. **Provide context**: Give relevant background information
3. **Ask for steps**: Request step-by-step explanations when needed
4. **Use appropriate models**: Match model capability to task complexity
5. **Monitor costs**: Reasoning models use more tokens
