Mistral Large 4 open weights land with a 41% terminal benchmark score
Mistral just released the open weights for Mistral Large 4 (ML4), its 1-trillion-parameter multimodal model, also known internally as Le Chonk. The company says three weeks of reinforcement learning after the initial launch produced a meaningful jump in agentic coding performance.
The Benchmark Numbers
ML4 now tops the Artificial Analysis Coding Agent Index, beating Z.ai’s GLM-5.3 while using 47% fewer tokens per task on average. On Terminal-Bench 4.0, which tests agentic coding in the terminal, ML4 climbed from 28% to 41%, tying GLM-5.3 for the top spot.
The model keeps 52 billion parameters active at any given time. Mistral’s VP of Science Pierre Stock attributes the improvement to two additional reasoning modes (low, medium, and high) added on top of the base model.
Why This Matters for Operators
Until now, the strongest open-weight coding models were developed in China. For teams with security concerns about that provenance, ML4 offers a French-developed alternative they can self-host on infrastructure they control or run through cloud providers that host Mistral models.
Mistral raised $6 billion to build its models, a smaller figure than most frontier labs. The weights are open, which means deployment is not locked to any vendor.
The Token Cost Angle
Stock also noted ML4 scores well on vulnerability discovery at a lower cost than rival frontier models. The tradeoff he flagged: open weights mean cybercriminals can access the same capability to develop exploits. That is a real consideration for any team weighing open versus closed models for security work.
Token costs are a growing concern as agentic tasks get more complex. A model that uses 47% fewer tokens per task is a meaningful variable in that calculation, especially for teams running coding agents at volume.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
