Mistral releases Mistral Large 4 (Le Chonk), a 1-trillion parameter open-weight model

Mistral Large 4 (Le Chonk) is a 1-trillion parameter native multimodal model released as fully open weights, trained from scratch on European infrastructure to support sovereign AI strategies.
The model leads the Cognition FrontierCode 2.0 benchmark at 19.7%, surpassing GLM-5.3 (19.3%) and Kimi K3 (17.8%), while using 47% fewer tokens per task than GLM-5.3 on the Artificial Analysis Coding Agent Index.
It achieves a 96.6% average refusal rate for malicious cyber requests, the highest among open models, and scores 51% on the Artificial Analysis Cyber Index.
Available via Mistral Studio, Mistral API, OpenRouter, Hugging Face, and Microsoft Foundry, with self-hosting support for vLLM and SGLang.

Performance Benchmarks

Terminal-Bench 4.0 score improved by 13 percentage points to 41% compared to the preview version, with a 30% reduction in output tokens.
FinWorkBench (Finch) score of 73.3% ranks second among open models, behind Kimi K3 (77.3%).
BrowseComp score of 84.5% exceeds models like GLM-5.3 and DeepSeek V4.1 Flash 1017.

Model weights allow organizations to inspect, customize, and deploy on their own infrastructure without relying on Mistral's API.
Mistral Forge enables customization with proprietary data, while Mistral Studio supports agent and application governance.
NVIDIA integration includes optimized NIM containers for vLLM serving and the NeMo AutoModel library for customization.

Released under the Mistral Ecosystem Open-Weight license (modified MIT), allowing free use, modification, and commercialization.
Businesses operating a Model-as-a-Service (MaaS) offering must obtain a separate commercial license from Mistral.
API pricing is $0.70 per million input tokens, $0.07 per million cached input tokens, and $2.10 per million output tokens.