# Grok 3 Mini
Streamlined version of Grok-3 optimized for quicker response times.
## Model Details
| Property | Value |
|----------|-------|
| Model ID | `grok-3-mini` |
| Context Length | 131,072 tokens (128K input + 8K max output) |
| Knowledge Cutoff | November 2024 |
| Input Modalities | Text, Images |
| Output Modality | Text |
| Release Date | February 17, 2025 |
## Pricing
| Type | Cost per 1M tokens |
|------|-------------------|
| Input | $0.30 |
| Output | $0.50 |
## Capabilities
- Text and image input support
- Function calling
- Structured outputs
- Reasoning with configurable effort
- Supports `reasoning_effort` parameter
## Supported Parameters
- `temperature`
- `top_p`
- `max_tokens`
- `tools`
- `tool_choice`
- `response_format`
- `seed`
- `logprobs`
- `top_logprobs`
- `reasoning_effort` (low/high)
## Reasoning Effort
Grok-3-mini supports the `reasoning_effort` parameter:
- `low` - Minimal thinking time, fewer tokens, quick responses
- `high` - Maximum thinking time, more tokens, complex problems
```python
response = client.chat.completions.create(
model="grok-3-mini",
messages=[{"role": "user", "content": "Solve this math problem"}],
reasoning_effort="high"
)
```
## Benchmarks
| Benchmark | Score |
|-----------|-------|
| AIME 2024 | 95.8% |
| AIME 2025 | 90.8% |
| GPQA | 84% |
| LiveCodeBench | 80.4% |
## Usage
```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-3-mini",
messages=[
{"role": "user", "content": "Quick question about Python"}
]
)
```
## Notes
- Only model that supports `reasoning_effort` parameter
- 10x cheaper than grok-3 for input tokens
- 30x cheaper than grok-3 for output tokens
- Ideal for high-volume, cost-sensitive applications
