Illustration, AI-generated from the story text. Tap a panel to enlarge.

Mistral posts the weights for its trillion-parameter Large 4

Mistral released the weights of Mistral Large 4, a mixture-of-experts model with about one trillion parameters and roughly 52 billion active per token, on Hugging Face in FP8 and NVFP4 builds. Mistral previewed the model earlier this month. The license is a modified MIT: anyone may use, fine-tune and sell on top of it, but a company selling the model itself as a hosted service needs a separate commercial license A.

Mistral's own numbers put it at 48% on Artificial Analysis's Coding Agent Index, level with Zhipu's GLM-5.3 while using 47% fewer tokens per task, 41% on Terminal-Bench 4.0, and 51% on Artificial Analysis's cyber index. It is pitched at defenders: Mistral reports it refuses 96.6% of malicious cyber requests and markets it as "sovereign cyber defense" for governments that want to run it themselves. The API costs $0.70 per million input tokens and $2.10 per million output A·B.

The comparison Mistral chose is a Chinese model, which is where the open-weight frontier has sat most of this year. SourcesABB

Sources

• AMistral Large 4 (Le Chonk) is now open weight (Mistral)mistral.ai
• BWelcome Le Chonk: Mistral Large 4 on Hugging Face (Hugging Face)huggingface.co
• BMistral releases weights for its trillion-parameter AI model (TechCrunch)techcrunch.com

This story ran in the Afternoon Brief of October 29th 2026.