Mistral AI Releases Leanstral: First Lean 4 Open Source Code Agent Capable of Automatically Outputting Formal Proofs

Gate News reports that on March 17, Mistral AI announced the release of Leanstral, the first open-source code agent specifically designed for the formal verification tool Lean 4. The core innovation of this model is that it can generate code and output formal proofs that can be automatically verified by Lean 4, bypassing the traditional bottleneck of manual review in AI code generation.

Leanstral uses a sparse MoE architecture, with a total of 120 billion parameters and 6 billion active parameters. It is open-sourced under the Apache 2.0 license and has been specially trained and optimized for lean-lsp-mcp. Users can launch it with zero configuration via the /leanstall command in Mistral Vibe or access it through the free API endpoint labs-leanstral-2603, and it also supports downloading weights for self-deployment.

Mistral also released a new evaluation benchmark, FLTEval, using the formalization of Fermat’s Last Theorem in the Lean 4 community as the test scenario. Performance comparisons show that Leanstral pass@2 scores 26.3 at a cost of $36, surpassing Claude Sonnet 4.6, which costs $549 and scores 23.7. For pass@16, it scores 31.9 at a cost of $290, leading Sonnet by 8 points, while Claude Opus 4.6 requires $1,650 to reach 39.6. Among open-source models, Qwen3.5-397B-A17B needs four runs to reach 25.4, still below Leanstral’s pass@2 performance.

Disclaimer: The information on this page may come from third-party sources and is for reference only. It does not represent the views or opinions of Gate and does not constitute any financial, investment, or legal advice. Virtual asset trading involves high risk. Please do not rely solely on the information on this page when making decisions. For details, see the Disclaimer.
Comment
0/400
No comments