A free, specialist model for mathematicians and engineers doing formal proofs in Lean 4, and of little use to anyone else.
Kenning's take
What it does
Leanstral 1.5 is a Mistral AI model trained to write proofs in Lean 4, a language where a computer checks every step of a mathematical proof. Mistral reports that it solves all miniF2F problems and 587 of 672 PutnamBench problems, and that it found 5 previously unknown bugs while verifying code in open-source repositories. The weights are free under Apache 2.0 on Hugging Face, and it is also a free API endpoint that Mistral recommends using through its Mistral Vibe coding agent.
- 119B total parameters, 6B active
- Apache 2.0 open weights on Hugging Face
- Free API endpoint named leanstral-1-5
- Agent mode in Mistral Vibe for proof engineering
- Works with the Lean language server through MCP
Who it suits. Mathematicians, researchers and engineers doing formal verification in Lean 4.
Key facts
- Category
- AI models
- Made by
- Mistral AI
- Released
- 2 July 2026
- How to get it
- free open weights on Hugging Face and a free Mistral API endpoint (an API key is needed); recommended inside Mistral Vibe
- Works on
- API, self-hosted, Mistral Vibe
- In the UK
- Yes
- Free plan
- free API endpoint and free open weights
- API price
- free endpoint at launch
FAQ
When was Leanstral 1.5 released?
Mistral AI released Leanstral 1.5 on 2 July 2026.
Is Leanstral 1.5 free?
Yes. The weights are Apache 2.0 on Hugging Face, and Mistral offers a free API endpoint.
How do I use Leanstral 1.5?
Mistral recommends installing Mistral Vibe, adding Leanstral with the /leanstall command and launching the lean agent.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools

ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.

Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.

Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.