Model Overview
Grok-3-Mini is a lightweight reasoning model that thinks before responding, featuring accessible raw thinking traces for transparency in logic-based tasks.
Key Features
- High intelligence (3/4 dots rating)
- Fast speed (4/5 lightning bolts rating)
- 131,072 context window
- Medium max output tokens (estimated 4,096+)
- November 17, 2024 knowledge cutoff
- Text input support
- Text output with thinking traces
Technical Specifications
- Pricing: $0.30 per 1M tokens (input), $0.50 per 1M tokens (output)
- Supports: Input: text; Output: text with reasoning traces
- Features: Chain-of-thought reasoning, accessible thinking traces, cost-effective
Snapshots
- grok-3-mini (alias for grok-3-mini-latest)
- grok-3-mini-latest
Positioning and Use Cases
Grok-3-Mini is designed for logic-based tasks that do not require deep domain knowledge. It excels at mathematical reasoning, logical puzzles, and step-by-step problem solving. The model's unique feature is its accessible raw thinking traces, making it perfect for educational applications, debugging logical processes, and scenarios where transparency in reasoning is important. It offers excellent value for cost-conscious applications requiring reasoning capabilities.
Rate Limits
- Information not publicly available
Additional Notes
- Knowledge Cutoff: All Grok-3 family models have a knowledge cutoff of November 17, 2024
- No Internet Access: Unlike grok.com and Grok in X, API models are not connected to the internet
- Flexible Role Order: No role order limitation - you can mix system, user, or assistant roles in any sequence
- Model Aliases: Latest versions are automatically updated through aliases for seamless upgrades
- Fast vs Standard: Fast variants offer identical quality with reduced latency at higher cost
Documentation
Official Documentation
