Mistral AI's Math Tool: 119B Parameters, Only 6B Active, Costs 1% of Rivals
European AI company Mistral AI has unveiled Leanstral 1.5, a model designed for formal mathematical proofs using the Lean4 programming language. Despite its massive 119 billion parameters, the model activates only 6 billion during inference, delivering impressive results at a fraction of the computational cost. And it's completely open-source under the Apache-2.0 license.

On core benchmarks, Leanstral 1.5 shines. It achieved a perfect 100% completion rate on both the validation and test sets of the miniF2F formal mathematics benchmark. On PutnamBench, a math competition dataset, it solved 587 out of 672 Lean4 problems. For the FATE series benchmarks in abstract algebra, it scored 87% on master's-level FATE-H and 34% on doctoral-level FATE-X—both best-in-class.
But the real head-turner is the cost. On PutnamBench, Leanstral 1.5 averages just $4 per problem. Compare that to ByteDance's Seed-Prover 1.5, which costs over $300 per problem, or Aleph Prover at $54–$68. That's roughly 1% of the strongest competitor's cost, removing a major economic barrier for large-scale formal proof applications.
In practical engineering tests, Leanstral 1.5 proved its worth. It identified 47 violating attributes across 57 code repositories, including 11 pointing to real code defects—5 of which were completely new issues never reported on GitHub. From pure math competitions to real-world software verification, this model shows that parameter size isn't everything; efficient activation is key to making AI reasoning practical.
Key Points
- Model: Leanstral 1.5, specialized for Lean4 formal proofs.
- Architecture: 119B total parameters, only 6B activated during inference.
- Performance: 100% on miniF2F; 587/672 on PutnamBench; top scores on FATE benchmarks.
- Cost: ~$4 per problem on PutnamBench, about 1% of competitors' costs.
- Practical Use: Found 47 violations in code repositories, including 5 previously unknown defects.
- License: Open-source under Apache-2.0.