Open Source

Mistral's Leanstral 1.5 achieves state-of-the-art formal verification with 6B active parameters

Open-source model solves 87% of FATE-H and uncovers 5 real-world bugs.

Deep Dive

Mistral released Leanstral 1.5, a free Apache-2.0 licensed model with 6B active parameters. It delivers a major performance upgrade in formal verification, saturating miniF2F, solving 587/672 PutnamBench problems, and achieving state-of-the-art results on FATE-H (87%) and FATE-X (34%). Trained through mid-training, supervised fine-tuning, and reinforcement learning with CISPO, it excels in agentic proof engineering and real-world code verification, uncovering 5 previously unknown bugs across 57 repositories tested. Leanstral 1.5 can be used for automated theorem proving and formal proof engineering, allowing developers to verify the correctness of their software and code specifications.

Key Points
  • Apache-2.0 licensed model with 119B total parameters, only 6B active at inference time.
  • Achieves SOTA on formal verification: 87% on FATE-H, 587/672 PutnamBench, and saturates miniF2F.
  • Found 5 real-world bugs across 57 software repositories during testing.

Why It Matters

Enables developers to automatically verify software correctness, reducing bugs and improving reliability.

📬 Get the top 10 AI stories daily