Mistral's open-source Leanstral 1.5 aces formal math benchmarks and catches real bugs in code

Formal verification has always been the corner of AI research that promises the most trustworthy code and gets the least attention, since proving code correct is much harder than making it look plausible. A free, open model that can both ace olympiad level proofs and catch real bugs in the wild is a signal that provably correct software might stop being a niche academic exercise. The interesting part is not the benchmark score but the bug hunting, since that is the use case regular developers could actually adopt without learning a proof language themselves.