Mistral releases Le Chonk ML4, an adaptive open coding model
Mistral just dropped Le Chonk (ML4), its latest open model for coding, and it's already making waves.
After three weeks of testing with select partners, ML4 is being called one of the top performers out there.
It also lets you pick between low, medium, and high reasoning modes, so whether you're debugging or building something from scratch, it adapts to your needs.
ML4 gets 41% on Terminal-Bench 4.0
ML4 scored big on the Terminal-Bench 4.0 test, jumping from 28% to 41%, which puts it ahead of Opencode's GLM-5.3 on Terminal-Bench 4.0 and right up there with Codex GPT-6.1 Terra on the Artificial Analysis Coding Agent Index.
Even better: it uses 47% fewer tokens per task than its competitors, so it's more efficient for users.
Mistral joins Open Secure AI Alliance
ML4 also did well on cybersecurity tests like CyberGym.
To show it's serious about safe AI, Mistral has joined NVIDIA's Open Secure AI Alliance, highlighting how open models can help tackle security challenges in AI today.