Model WarsJuly 3, 2026via MarkTechPost
Mistral AI Releases Leanstral 1.5: An Apache-2.0 Lean 4 Code Agent Model Solving 587 of 672 PutnamBench Problems
Why it matters
Mistral ships a specialized code agent model that outperforms on formal reasoning benchmarks while staying fully open source. This signals a shift toward task-specific, efficient MoE models competing with closed generalists on narrow but valuable domains.
Key signals
- Leanstral 1.5 solves 587 of 672 PutnamBench problems
- 119B parameter mixture-of-experts model activating 6.5B parameters per token
- Apache 2.0 license (fully open source)
- Saturates miniF2F benchmark
- Specialized for Lean 4 formal verification and code generation
- Includes bug-finding case studies and deployment code
The hook
587 of 672. That's Mistral's new Leanstral 1.5 solving Putnam-level math proofs—open source, Apache 2.0, and it's already finding real bugs.
Mistral AI released Leanstral 1.5, a free Apache-2.0 code agent model for Lean 4. It saturates miniF2F and solves 587 of 672 PutnamBench problems. The 119B mixture-of-experts activates 6.5B parameters per token. We break down its architecture, benchmarks, real bug-finding case studies, and deployment code.