From SiliconANGLE · Kyt Dotson · · 1m 10s
Mistral releases open weights for Mistral Large 4
Mistral AI opens weights for its largest model, Le Chonk, to give companies sovereign control over AI.
In brief
Mistral AI opens weights for its largest model, Le Chonk, to give companies sovereign control over AI. Mistral AI released open weights for Mistral Large 4, its most capable model, to let companies deploy it on their own hardware without cloud exposure. Originally reported by SiliconANGLE.
Mistral releases open weights
Mistral AI SAS announced open weights for its frontier model Mistral Large 4, nicknamed Le Chonk.
Model specs
ML4 has 1 trillion parameters and 52 billion active parameters, natively multimodal, and received three weeks of reinforcement training.
Performance claims
ML4 is one of the best open agentic coding models globally, using 47% fewer tokens per task than top models.
Training details
Mistral trained ML4 on 3,800 Nvidia Grace Blackwell GPUs in European data centers for high-fidelity tool use and coding.
Sovereign control
Companies can deploy ML4 on their own hardware within firewalls without exposing sensitive data to the cloud.
Competitive context
Mistral’s model competes with Chinese coding models like GLM-5.3 from Zhipu and Kimi from Moonshot AI.
Reasoning modes
ML4 offers low, medium, and high reasoning modes to match task complexity without adjusting prompts.
The key point
Mistral AI released open weights for Mistral Large 4, its most capable model, to let companies deploy it on their own hardware without cloud exposure.
