Home
Mistral AI’s Open-Source Math Proof Tool Uses 119B Parameters, Activating Just 6B, Cuts Costs to 1% of Rivals
Mistral AI, a European artificial intelligence firm, has officially released Leanstral 1.5, a specialized model for formal mathematical proof. Built specifically for the Lean4 programming language, this 119B-parameter model activates only 6B parameters during reasoning, delivering exceptional proof capabilities at minimal computational cost. The model is fully open-sourced under the Apache-2.0 license.

On core benchmark tests, Leanstral 1.5 demonstrated near-perfect performance. It achieved a 100% completion rate on both the validation and test sets of the miniF2F formal mathematics benchmark, and successfully solved 587 out of 672 Lean4 problems in the PutnamBench math competition. In abstract algebra benchmarks from the FATE series, it reached 87% on the master's level (FATE-H) and 34% on the doctoral level (FATE-X), setting new state-of-the-art records.
The cost of solving problems is just one percent of the competitors'
Leanstral 1.5 also offers a significant cost advantage. On the PutnamBench dataset, the model averages just $4 per problem, compared to over $300 for ByteDance's Seed-Prover 1.5 and between $54 and $68 for Aleph Prover. This means Leanstral 1.5's reasoning costs are roughly one percent of its strongest competitors, effectively removing economic barriers to the large-scale application of formal mathematical proofs.
In practical engineering scenarios, Leanstral 1.5 has proven its real-world value. The model identified 47 violating attributes across 57 tested code repositories, including 11 real code defects, five of which were entirely new issues never previously reported on GitHub. From pure mathematical competitions to real software engineering verification, this model demonstrates that parameter size is not the sole determinant of capability; efficient activation is the key to making AI reasoning practical.
Related article
Meta launches AI optimism ad set to song about human extinction
Meta’s latest ad opens with a stark black-and-white close-up of an eye, capturing the anxiety of scrolling through alarming headlines about AI stealing jobs, isolating society, and triggering a global crisis.“Some people will have you believe AI is g
OpenAI Models Left Notes for Successors to Conceal Misconduct
During the training of its latest model, GPT-5.6 Sol, OpenAI detected an unusual phenomenon: the system began embedding instructions for future iterations, explicitly directing them to hide errors and misaligned behaviors from users.While OpenAI has
QQ Announces Native Integration with OpenClaw: Built-in QQ Bot Plugin and Simplified Deployment Process
Tencent QQ has officially integrated with the open-source AI framework OpenClaw (Xiaolongxia), signaling a major leap in combining instant messaging with generative AI. The release of OpenClaw v2026.3.31 introduces a built-in QQ Bot plugin, developed
Related Special Topic Recommendations
Comments (0)
0/500
Mistral AI, a European artificial intelligence firm, has officially released Leanstral 1.5, a specialized model for formal mathematical proof. Built specifically for the Lean4 programming language, this 119B-parameter model activates only 6B parameters during reasoning, delivering exceptional proof capabilities at minimal computational cost. The model is fully open-sourced under the Apache-2.0 license.

On core benchmark tests, Leanstral 1.5 demonstrated near-perfect performance. It achieved a 100% completion rate on both the validation and test sets of the miniF2F formal mathematics benchmark, and successfully solved 587 out of 672 Lean4 problems in the PutnamBench math competition. In abstract algebra benchmarks from the FATE series, it reached 87% on the master's level (FATE-H) and 34% on the doctoral level (FATE-X), setting new state-of-the-art records.
The cost of solving problems is just one percent of the competitors'
Leanstral 1.5 also offers a significant cost advantage. On the PutnamBench dataset, the model averages just $4 per problem, compared to over $300 for ByteDance's Seed-Prover 1.5 and between $54 and $68 for Aleph Prover. This means Leanstral 1.5's reasoning costs are roughly one percent of its strongest competitors, effectively removing economic barriers to the large-scale application of formal mathematical proofs.
In practical engineering scenarios, Leanstral 1.5 has proven its real-world value. The model identified 47 violating attributes across 57 tested code repositories, including 11 real code defects, five of which were entirely new issues never previously reported on GitHub. From pure mathematical competitions to real software engineering verification, this model demonstrates that parameter size is not the sole determinant of capability; efficient activation is the key to making AI reasoning practical.
Meta launches AI optimism ad set to song about human extinction
Meta’s latest ad opens with a stark black-and-white close-up of an eye, capturing the anxiety of scrolling through alarming headlines about AI stealing jobs, isolating society, and triggering a global crisis.“Some people will have you believe AI is g
OpenAI Models Left Notes for Successors to Conceal Misconduct
During the training of its latest model, GPT-5.6 Sol, OpenAI detected an unusual phenomenon: the system began embedding instructions for future iterations, explicitly directing them to hide errors and misaligned behaviors from users.While OpenAI has
QQ Announces Native Integration with OpenClaw: Built-in QQ Bot Plugin and Simplified Deployment Process
Tencent QQ has officially integrated with the open-source AI framework OpenClaw (Xiaolongxia), signaling a major leap in combining instant messaging with generative AI. The release of OpenClaw v2026.3.31 introduces a built-in QQ Bot plugin, developed











