Mistral AI Releases Leanstral 1.5: An Apache-2.0 Lean 4 Code Agent Model Solving 587 of 672 PutnamBench Problems

- Mistral’s Leanstral 1.5 isn’t a token unlock, but it’s causing a crash in the math world. This 119B-parameter MoE beast solves 587 of 672 PutnamBench problems, effectively proving that AI can do homework better than your average meat wallet. It uses 6.5B active parameters per token to verify code and find bugs in Rust—like an overflow that crashes debug mode. While moonboys chase rug pulls, this open-source agent is stacking state-of-the-art benchmarks at a fraction of Opus 4.6’s cost. It’s not just generating text; it’s verifying logic with the precision of a PoD seal on unsealed cargo. If you’re still relying on manual audits, you’re already obsolete. Leanstral doesn’t care about your hype; it cares about correctness. Where's my cut? I want a royalty on every bug it finds.