Leanstral 1.5 : l’abondance de preuves pour tous

Leanstral 1.5 is a new, freely available AI model that dramatically improves performance in formal mathematical proof and code verification. The model demonstrates remarkable capabilities by achieving state-of-the-art results on complex benchmarks, such as solving a high percentage of problems from the Putnam Mathematical Competition. Its advanced skill comes from a unique training process involving reinforcement learning, where it learns to operate like a developer by modifying files and using compiler feedback to refine proofs. In practical applications, Leanstral 1.5 has successfully verified intricate code properties and even discovered previously unknown bugs in real-world open-source software.