Overview
Mistral released Mistral Small 4, a new 119B parameter model that unifies reasoning, multimodal, and coding capabilities into a single versatile model. The Apache 2 licensed model introduces variable reasoning effort settings and comes alongside Leanstral, a specialized model for formal mathematical proof verification.
Key Facts
- 119B parameter Mixture-of-Experts architecture with 6B active parameters - enables powerful AI capabilities while maintaining efficiency
- Unifies Magistral (reasoning), Pixtral (multimodal), and Devstral (coding) into one model - eliminates need for multiple specialized models
- Apache 2 open source license - allows commercial use and modification without restrictions
- Variable reasoning effort settings (none/high) - users can choose between speed and thoroughness
- Released alongside Leanstral for Lean 4 formal verification - AI can now help with mathematically provable code
Why It Matters
This represents a significant consolidation in AI model architecture where unified models replace specialized tool chains, potentially simplifying AI deployment while maintaining performance across diverse tasks.