Mistral released Leanstral 1.5, a code model that goes beyond generating source code to also generating and checking formal mathematical proofs of correctness, written in the Lean 4 proof assistant language. Where a normal test suite can only show that code passes the cases you thought to write, a Lean 4 proof establishes a property of the code mathematically -- the kind of guarantee that aviation software, medical device firmware, and cryptographic libraries have historically required, at the cost of enormous manual labor from formal-methods specialists. Leanstral automates both ends of that pipeline: translating a specification into Lean 4 statements (autoformalization) and then searching for and verifying a proof that the generated code satisfies them. For developers, the practical implication is that formal verification, previously reserved for the highest-budget safety-critical projects, becomes something ordinary teams can turn on for a specific module -- payment processing, authentication, a scheduler used in a medical device -- without hiring a dedicated proof engineer. It is also a pointed response to a real problem: as more production code is AI-generated, the tests that accompany it are often AI-generated too and can share the same blind spots as the code they check, so a verification method that does not depend on the model's own judgment about what to test is a meaningfully different quality bar. Mistral has been building a portfolio of narrow, high-value technical capabilities -- it shipped a single-camera robot navigation model the week before -- rather than chasing frontier general-purpose benchmarks outright, and Leanstral fits that pattern of owning a hard, unglamorous problem the larger labs treat as a side project.