# Grok 3
xAI's advanced reasoning model with exceptional benchmark performance.
## Model Details
| Property | Value |
|----------|-------|
| Model ID | `grok-3` |
| 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 | $3.00 |
| Output | $15.00 |
## Capabilities
- Text and image input support
- Function calling
- Structured outputs
- Reasoning enabled
- Real-time web search capable
## Supported Parameters
- `temperature`
- `top_p`
- `max_tokens`
- `tools`
- `tool_choice`
- `response_format`
- `seed`
- `logprobs`
- `top_logprobs`
## Benchmarks
| Benchmark | Score |
|-----------|-------|
| AIME 2025 | 93.3% |
| GPQA | 84.6% |
| LiveCodeBench | 79.4% |
| MMMU | 78.0% |
## 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",
messages=[
{"role": "user", "content": "Explain quantum computing"}
]
)
```
## Notes
- Ten times more computational resources than Grok-2
- Does not support `reasoning_effort` parameter
- Use `grok-3-mini` for faster, more economical responses
