Leanstral 1.5: Proof abundance for all
mistral.ai
July 3, 2026
6 min read
🔥🔥🔥🔥🔥
64/100
Summary
Leanstral 1.5 is a free Apache-2.0 licensed model with 6 billion active parameters that significantly enhances performance in formal verification. It solves 587 out of 672 PutnamBench problems, achieves state-of-the-art results on FATE-H (87%) and FATE-X (34%), and uncovers five previously unknown bugs across 57 repositories.
Key Takeaways
- Leanstral 1.5 is a free Apache-2.0 licensed model with 6B active parameters that significantly enhances performance in formal verification tasks.
- The model solves 587 out of 672 problems on the PutnamBench and achieves state-of-the-art results of 87% on FATE-H and 34% on FATE-X.
- Leanstral 1.5 uncovers five previously unknown bugs across 57 tested repositories, demonstrating its effectiveness in real-world code verification.
- The model was trained using mid-training, supervised fine-tuning, and reinforcement learning with CISPO, allowing it to excel in proof engineering workflows.
Coverage & discussions
Hacker News · mistral.ai
Community discussions
Hacker News discussion ↗
Historical count: 5 · 5 saved in sample · Historical sample
Analysis ready
Community Sentiment
Positives
- Leanstral 1.5 is making waves with its formal verification claims, showcasing the potential for proving the absence of bugs rather than just finding them.
- The integration with OpenATP is a game-changer, allowing users to leverage automated theorem provers in a flexible environment.
- The excitement about 'frontier small language models' indicates a growing interest in compact yet powerful AI solutions.
- Lean is gaining traction as a functional programming language, which could enhance its adoption in software verification compared to traditional tools.
Concerns
- Critics are skeptical about the model's claimed improvements, pointing out that comparisons are made with outdated models, which feels like a cheap shot.
- Concerns about the effectiveness of testing and fuzzing are raised, suggesting that the marketing may oversell the model's capabilities.
- Commenters question the credibility of the authors, implying that missing basic edge cases reflects poorly on their expertise.
- There’s a sense that Lean's adoption in formal verification is still lagging behind more established systems, raising doubts about its viability.
