[{"data":1,"prerenderedAt":30},["ShallowReactive",2],{"nr-en-mistral-leanstral-formale-mathematik-code-verifikation":3},{"slug":4,"title":5,"dek":6,"date":7,"time":8,"publishedAt":9,"updated":10,"updatedAt":10,"dateFmt":11,"updatedFmt":10,"kind":12,"tier":13,"author":14,"authorName":15,"topics":16,"tracker":22,"trackerLabel":23,"headlineStat":24,"image":25,"ogImage":26,"imageAlt":5,"csv":10,"minutes":27,"words":28,"html":29},"mistral-leanstral-formale-mathematik-code-verifikation","Mistral Releases Leanstral 1.5: Open-Source Model Finds Real Bugs in Code","Mistral AI's new model solves complex mathematical proofs and discovers security vulnerabilities in real code. In a scan of 57 open-source projects, it already found five previously unknown bugs.","2026-07-04","15:07","2026-07-04T15:07:00+02:00","","July 4, 2026","news","standard","ideal-syka","Ideal Syka",[17,18,19,20,21],"Open-Source AI","Formal Verification","Code Security","Mathematics AI","Mistral AI","\u002Flokale-ki","Local AI & Open Source","587 of 672 tasks solved in Putnam Benchmark; 5 real bugs found in 57 repositories","\u002Fnewsroom\u002Fimg\u002Fmistral-leanstral-formale-mathematik-code-verifikation.webp","\u002Fog-nr\u002Fmistral-leanstral-formale-mathematik-code-verifikation.en.png",2,496,"\u003Cp>Mistral AI has released \u003Cstrong>Leanstral 1.5\u003C\u002Fstrong>, a free open-source model under Apache 2.0 license, specialized in formal verification using the Lean 4 programming language. The key distinction: the model not only masters mathematical proofs but also finds real bugs in production code – a concrete breakthrough in AI-assisted software verification.\u003C\u002Fp>\n\u003Ch2>The essentials\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Cstrong>Leanstral 1.5\u003C\u002Fstrong> achieves \u003Cstrong>100 percent\u003C\u002Fstrong> on the miniF2F benchmark for formal mathematics and solves \u003Cstrong>587 of 672\u003C\u002Fstrong> tasks in the demanding PutnamBench\u003C\u002Fli>\n\u003Cli>When scanning \u003Cstrong>57 open-source repositories\u003C\u002Fstrong>, the model found \u003Cstrong>five previously unknown bugs\u003C\u002Fstrong>, including an \u003Cstrong>overflow bug in the Rust library varinteger\u003C\u002Fstrong>\u003C\u002Fli>\n\u003Cli>The model is free via \u003Cstrong>Hugging Face\u003C\u002Fstrong> and a \u003Cstrong>free API\u003C\u002Fstrong>\u003C\u002Fli>\n\u003Cli>Training used Mid-Training, supervised fine-tuning, and \u003Cstrong>Reinforcement Learning\u003C\u002Fstrong>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>Mathematics at olympiad level\u003C\u002Fh2>\n\u003Cp>The benchmark results demonstrate the model's reach: on miniF2F, which covers tasks from school level to olympiad problems, Leanstral 1.5 achieves a \u003Cstrong>perfect score of 100 percent\u003C\u002Fstrong>. In \u003Cstrong>PutnamBench\u003C\u002Fstrong> – a benchmark with 672 problems from the renowned Putnam Mathematics Competition – it solves \u003Cstrong>587 problems\u003C\u002Fstrong>. On the more demanding algebra benchmarks FATE-H and FATE-X, which test master's and doctoral-level problems in areas like group theory and ring theory, the model achieves \u003Cstrong>87 and 34 percent\u003C\u002Fstrong> respectively.\u003C\u002Fp>\n\u003Cdiv class=\"tbl-scroll\">\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Benchmark\u003C\u002Fth>\n\u003Cth>Tasks\u003C\u002Fth>\n\u003Cth>Leanstral 1.5\u003C\u002Fth>\n\u003Cth>Characteristic\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\u003Ctr>\n\u003Ctd>miniF2F\u003C\u002Ftd>\n\u003Ctd>variable\u003C\u002Ftd>\n\u003Ctd>100 %\u003C\u002Ftd>\n\u003Ctd>School to olympiad level\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>PutnamBench\u003C\u002Ftd>\n\u003Ctd>672\u003C\u002Ftd>\n\u003Ctd>587 solved\u003C\u002Ftd>\n\u003Ctd>Renowned math competition\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>FATE-H\u003C\u002Ftd>\n\u003Ctd>Master level\u003C\u002Ftd>\n\u003Ctd>87 %\u003C\u002Ftd>\n\u003Ctd>Group theory, ring theory\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>FATE-X\u003C\u002Ftd>\n\u003Ctd>Doctoral level\u003C\u002Ftd>\n\u003Ctd>34 %\u003C\u002Ftd>\n\u003Ctd>Highest difficulty\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2>From math to real bugs\u003C\u002Fh2>\n\u003Cp>What makes the model exceptional: although trained primarily on mathematical verification, it demonstrates strong capabilities in practical \u003Cstrong>code verification\u003C\u002Fstrong> according to Mistral. This isn't just theory – when scanning 57 open-source repositories, Leanstral 1.5 found \u003Cstrong>five previously unknown bugs\u003C\u002Fstrong>. Among them was an \u003Cstrong>overflow bug in the Rust library varinteger\u003C\u002Fstrong>, a genuine security risk lurking in production code.\u003C\u002Fp>\n\u003Cp>This means: the model can not only formally verify mathematical proofs but also scan real software projects for errors – and finds things previous methods missed.\u003C\u002Fp>\n\u003Ch2>Training with three techniques\u003C\u002Fh2>\n\u003Cp>The performance builds on a combined training approach: \u003Cstrong>Mid-Training\u003C\u002Fstrong>, \u003Cstrong>supervised fine-tuning\u003C\u002Fstrong>, and \u003Cstrong>Reinforcement Learning\u003C\u002Fstrong> were used together to optimize the model for both mathematical precision and practical code analysis.\u003C\u002Fp>\n\u003Ch2>What this means for enterprises\u003C\u002Fh2>\n\u003Cp>For software developers and security teams, Leanstral 1.5 could become relevant for code review and bug detection. A free, open-source model that finds real bugs could become part of a local AI infrastructure – without dependency on external APIs. However, questions remain about how reliable the model is with larger, more complex codebases and how it performs in compliance scenarios. But the release demonstrates: formal verification with AI is no longer a future prospect but practically deployable today.\u003C\u002Fp>\n\u003Ch2>Sources\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fthe-decoder.de\u002Fmistral-veroeffentlicht-open-source-modell-fuer-formale-mathematik-und-code-verifikation\u002F\">The Decoder (DE)\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fthe-decoder.com\u002Fmistrals-open-source-leanstral-1-5-aces-formal-math-benchmarks-and-catches-real-bugs-in-code\u002F\">The Decoder\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cem>Editorially owned by \u003Ca href=\"\u002Fen\u002Fautor\u002Fideal-syka\">Ideal Syka\u003C\u002Fa>. Sources and method: \u003Ca href=\"\u002Fen\u002Fredaktion\">Newsroom & method\u003C\u002Fa>. Tips and corrections: \u003Ca href=\"mailto:ai@i6eal.de\">ai@i6eal.de\u003C\u002Fa>.\u003C\u002Fem>\u003C\u002Fp>\n",1783276596548]