Mistral AI Releases Leanstral 1.5: An Apache-2.0 Lean 4 Code Agent Model Solving 587 of 672 PutnamBench Problems
Today, Mistral AI released Leanstral 1.5. It is a code agent model built for Lean 4. The release targets automated theorem proving and proof engineering. Weights are open under Apache 2.0. A free API endpoint, leanstral-1-5, is now live. Leanstral 1.5 updates the earlier Leanstral-2603 model. It belongs to the Mistral Small 4 family. What is Leanstral 1.5 Leanstral 1.5 is a code agent model for Lean 4, a proof assistant. A proof assistant checks every logical step mechanically. Lean 4
Make your prediction
Will Leanstral 1.5 appear on the Mistral AI Hugging Face page by July 8, 2026?
Resolves by Jul 8, 2026
Your prediction
50% · 50/50 coin flip
NOYES
Get smart on it
Mistral AI released Leanstral 1.5, a code agent model designed for Lean 4, a proof assistant that mechanically checks logical steps. The model uses a mixture-of-experts architecture with 119 billion total parameters and achieves strong performance on mathematical benchmarks, solving 587 of 672 problems on PutnamBench and reaching 100 percent on validation and test sets for another benchmark. The model was trained through three stages including reinforcement learning in environments where it attempts to prove theorems and work with code in a filesystem. The weights are released openly under Apache 2.0 license and can be accessed through a free API endpoint or self-hosted using specified software packages.
Related news
Leanstral 1.5: Proof Abundance for All
Leanstral 1.5 is a free, open-source AI model designed to automatically generate mathematical proofs and verify computer code using the Lean 4 formal verification system. The model was trained through three stages: mid-training, supervised fine-tuning, and reinforcement learning, allowing it to learn from compiler feedback when proofs fail and to work across multiple files like a developer would. It matters because formal verification can catch bugs and prove properties are correct, and this model makes that process more practical and affordable than existing alternatives, as demonstrated by discovering previously unknown bugs in real open-source repositories and solving complex mathematical problems at a fraction of the cost of competing systems.
Interfaze Ships diffusion-gemma-asr-small, an Open-Source Diffusion ASR Model Transcribing Six Languages via DiffusionGemma’s Parallel Denoising Decoder
Interfaze, a young YC’s startup, has open-sourced a new speech recognition model. It is called diffusion-gemma-asr-small. The model transcribes audio through a diffusion decoder, not an autoregressive one. It is described as the first multilingual audio diffusion ASR model. One adapter handles six languages. The research team trained only about 42M parameters on top of a frozen 26B backbone. That is roughly 0.16% of the model’s weights. Here two terms matter up front. Autoregress
