# Mistral Releases Leanstral 1.5 Formal Verification Model

> Mistral has released Leanstral 1.5, an Apache-licensed model built specifically for formal verification, the painstaking work of proving that mathematical statements and code actually hold. It is a 119-billion-parameter mixture-of-experts design that activates only about 6 billion parameters per token, and it posted a perfect 100 percent on both the validation and test splits of the miniF2F math benchmark. More striking than the benchmark is what happened when it was pointed at real software. Across 57 open-source repositories, Leanstral surfaced five previously unreported bugs, and on the harder PutnamBench set of 672 competition problems it solved 587. Formal methods have long been the rigorous but labor-intensive corner of computer science, mostly out of reach for everyday projects. A capable open model that can both prove theorems and catch genuine defects hints at a near future where verification is cheap enough to run routinely. That it comes from Mistral, under a permissive license, keeps this frontier from belonging solely to the largest US labs.

_Section: [Daily AI Updates](https://www.wortins.com/daily-ai) · Source: Mistral · Published Tuesday, July 21, 2026_

## Wortins' read

Mistral has released Leanstral 1.5, an Apache-licensed model built specifically for formal verification, the painstaking work of proving that mathematical statements and code actually hold. It is a 119-billion-parameter mixture-of-experts design that activates only about 6 billion parameters per token, and it posted a perfect 100 percent on both the validation and test splits of the miniF2F math benchmark. More striking than the benchmark is what happened when it was pointed at real software. Across 57 open-source repositories, Leanstral surfaced five previously unreported bugs, and on the harder PutnamBench set of 672 competition problems it solved 587. Formal methods have long been the rigorous but labor-intensive corner of computer science, mostly out of reach for everyday projects. A capable open model that can both prove theorems and catch genuine defects hints at a near future where verification is cheap enough to run routinely. That it comes from Mistral, under a permissive license, keeps this frontier from belonging solely to the largest US labs.

## Source

[Read the full story at Mistral](https://mistral.ai/news/leanstral-1-5/)

## Related coverage

- [OpenAI Launches GPT-5.6 in Three Tiers: Sol, Terra, Luna](https://www.wortins.com/story/openai-launches-gpt-5-6-in-three-tiers-sol-terra-luna-b5bae914) — [ThursdAI](https://thursdai.news/releases/2026-07)
- [ElevenLabs Moves v3 Out of Alpha with Emotion and Multi-Speaker Support](https://www.wortins.com/story/elevenlabs-moves-v3-out-of-alpha-with-emotion-and-multi-spea-c5541def) — [ElevenLabs](https://medium.com/the-ai-entrepreneurs/elevenlabs-in-2026-the-complete-guide-to-v3-agents-music-and-scribe-7f3c3bdfd201)
- [UN Global Dialogue Warns AI Could Cause Catastrophic Harm Without Global Governance](https://www.wortins.com/story/un-global-dialogue-warns-ai-could-cause-catastrophic-harm-wi-6d6bd188) — [UN News](https://news.un.org/en/story/2026/07/1167862)
- [xAI Releases Grok 4.5, Opus-Class Model Trained on Cursor Data](https://www.wortins.com/story/xai-releases-grok-4-5-opus-class-model-trained-on-cursor-dat-cee001a4) — [TechCrunch](https://techcrunch.com/2026/07/08/spacexai-releases-grok-4-5-which-elon-describes-as-an-opus-class-model/)
- [Thinking Machines Releases Inkling Open AI Model](https://www.wortins.com/story/thinking-machines-releases-inkling-open-ai-model-3293c584) — [Fortune](https://fortune.com/2026/07/15/what-is-mira-murati-thinking-machines-first-ai-model-inkling/)
- [Google Releases NanoBanana 2 Lite and OmniFlash Video Generation](https://www.wortins.com/story/google-releases-nanobanana-2-lite-and-omniflash-video-genera-ae180051) — [Google DeepMind](https://thursdai.news/releases/2026-07)

---

_Curated and written by [Wortins](https://www.wortins.com) — The daily AI briefing. Every story links to its original source; the "Wortins read" on each is our own original analysis. [About Wortins & our editorial approach](https://www.wortins.com/about)._