Mistral Releases Open Weights for Trillion-Parameter Mistral Large 4
By Lucas Ropek · · TexaHub · LLMs & Foundation Models, Model Comparisons, Regional Tech Ecosystems
Mistral published the weights of Mistral Large 4, a 1-trillion-parameter mixture-of-experts model, on Hugging Face on October 29, three weeks after an API preview. API pricing is $0.70 per million input tokens and $2.10 per million output tokens.
Weights on Hugging Face
French AI company Mistral released the weights of Mistral Large 4 on October 29, letting businesses and governments download and run the model on their own infrastructure. The model, nicknamed "Le Chonk", was first introduced as a public preview earlier in October. [1] [2]
Mistral describes the model as a natively multimodal, hybrid instruct-and-reasoning mixture-of-experts model with 1 trillion total parameters, about 52 billion of them active, and support for more than 160 languages. It offers low, medium and high reasoning modes, with high as the default, and was trained from scratch on European infrastructure. [1]
Mistral said Large 4 ranked ahead of GLM-5.3, Kimi K3 and DeepSeek V4.1 Flash in Surge's human-preference evaluation for code, behind only Anthropic's Claude Opus 5.5, and scored 19.7% on Cognition's FrontierCode 2.0, the highest among the open models it compared. [1]
Coding and Cyber Claims
Mistral said additional reinforcement learning since the preview raised its Terminal-Bench 4.0 score from 28% to 41% while cutting output tokens by 30%. It reported 48% on the Artificial Analysis Coding Agent Index, effectively tied with Chinese open model GLM-5.3 while using 47% fewer tokens per task, and called Large 4 the best open-weight model for coding. [1] [2]
On security, the company reported 51% on the Artificial Analysis Cyber Index, an average 96.6% refusal rate on malicious cyber requests and 97.1% attack resistance on the B3 Agent Security Benchmark, up from 93.3% in the preview. Booz Allen said in the announcement that the model matched vulnerability detection recall rates of other frontier cyber models. These figures are Mistral's own. [1]
Pierre Stock, Mistral's vice president of science, told TechCrunch that open weights provide access "that can't be cut off by definition", and said the company does not want to "fearmonger people" about AI security. [2]
The company framed the release around sovereign deployment, saying it "brings state-of-the-art sovereign cyber defense capabilities to enterprises and governments globally." Chief executive Arthur Mensch said AI should "be a tool for empowerment, not dominance." [2]
Licence and Pricing
The weights are published under the Mistral Ecosystem Open-Weight licence, a modified MIT licence that permits use, modification, fine-tuning and commercial deployment but requires a separate commercial licence for companies offering the model as a hosted service. [1]
The model is also available through the Mistral API and Mistral Studio, OpenRouter, Nvidia-hosted endpoints and NIM containers, and Microsoft Foundry, and can be served with vLLM, SGLang and TensorRT LLM. API pricing is $0.70 per million input tokens, $0.07 per million cached input tokens and $2.10 per million output tokens. [1]
Sources
[1] Mistral Large 4 (Le Chonk) is now open weight — Mistral AI
[2] Mistral releases weights for its trillion-parameter AI model, targeting cybersecurity — TechCrunch