Mistral Releases Leanstral 1.5 for Lean 4 Proof Work, Available Free in Labs

Mistral AI has announced the release of Leanstral 1.5, a significant update to its specialized AI model for formal proof engineering. This new version is tailored for automated theorem proving and autoformalization within the Lean 4 environment, a system crucial for mechanically verifying mathematical proofs and software correctness. The model leverages a sparse mixture-of-experts (MoE) architecture, boasting 119 billion total parameters with 6.5 billion active during processing, which contributes to its efficiency and ability to handle complex tasks. The core purpose of Leanstral 1.5 is to bridge the gap between human-readable mathematical and software specifications and the precise, machine-verifiable language required by Lean 4. Formal proofs, while offering unparalleled certainty, traditionally demand extensive manual effort to translate concepts into a format that Lean 4 can understand. Leanstral 1.5 aims to alleviate this burden, potentially expanding the application of formal verification beyond a niche group of specialists to broader practical use cases in software development and mathematics. An notable feature of Leanstral 1.5 is its substantial context window of 256k tokens, enabling it to process lengthy proof files and associated code comprehensively. This capacity is vital for maintaining coherence and accuracy across complex formal verification projects. The model is presented as an evolution of the original Leanstral, released in March 2026, which was noted for its ability to outperform larger open models in formal proofs at a fraction of the cost. Mistral AI is making Leanstral 1.5 accessible for free within its Labs tier, allowing developers and researchers to experiment with its capabilities through the console playground. This strategic move underscores Mistral's dedication to fostering innovation in the field of formal methods and making advanced AI tools more widely available. While the new model card does not yet include updated benchmark numbers against its predecessor or other competing systems, its release signifies continued progress in creating AI that can reliably assist in high-stakes verification tasks where even subtle errors can have significant consequences.