Product Launchmixture of expertsmultimodalmistrallean4
Mistral Releases Mistral Small 4 Model
8.7
March 16, 2026 — Mistral released Mistral Small 4, an Apache-2 licensed 119B-parameter Mixture-of-Experts model with a 6B active expert configuration that unifies reasoning, multimodal, and agentic coding capabilities. The model supports reasoning_effort="none" or "high", is available as a 242GB upload on Hugging Face, and Mistral also published Leanstral, a Lean 4–focused open-weight model.
Key Points
- 1Introduces 119B-parameter Mixture-of-Experts model with 6B active experts, Apache-2 licensed weights
- 2Unifies Magistral reasoning, Pixtral multimodal, and Devstral agentic coding capabilities into one versatile model
- 3Enables mode switching via reasoning_effort ('none'/'high') and offers 242GB Hugging Face download
Scoring Rationale
High-impact, actionable release with open MoE weights and multimodal/agentic consolidation; limited documentation and shallow initial coverage reduce certainty.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
