Mistral, the European AI flag bearer, has released its largest open weights model yet: Mistral Large 4, affectionately dubbed Le Chonk. The trillion-parameter large language model (LLM) uses a mixture-of-experts (MoE) architecture with about 49 billion active parameters, keeping serving costs relatively low. According to Mistral, Le Chonk was trained on approximately 3,800 Grace Blackwell GPUs (around 52 NVL72 racks) in the company's own European data centers, on a corpus spanning over 160 languages using supervised pre-training and reinforcement learning.
The model is multimodal and supports reasoning tasks. While Mistral did not specify hardware requirements, the model is small enough to run on 8-way GPU boxes such as Nvidia's HGX B300 or AMD's MI355X.
Early independent benchmarking by Artificial Analysis places Le Chonk's intelligence between DeepSeek V4.1 Flash and OpenAI's entry-level GPT6 Luna models. However, Mistral's model significantly outperforms Thinking Machines Lab's Inkling, the most capable open weights model from the United States. Mistral's own benchmarks — covering coding, agentic, finance, and legal tasks — show the model trading blows with top Chinese offerings from Alibaba, Moonshot, Z.AI, and DeepSeek.
Mistral is particularly highlighting Le Chonk's performance in cybersecurity workloads. The company claims it is one of the strongest performers on Artificial Analysis' new Cyber index. This positions the model against proprietary American models from OpenAI and Anthropic, which often refuse cybersecurity-related tasks that could be perceived as adversarial. "Provider-level refusals can block legitimate vulnerability research and incident response, and where losing access to a capability mid-incident can itself become a critical security risk. ML4 pairs top-tier cyber performance with open weights and self-deployment, giving organizations both the capability and the autonomy to run advanced security work under their own policies," Mistral wrote in a blog post.
Le Chonk is currently in preview, with its weights expected to be released on Hugging Face and other model repositories within the month. Mistral says this is the first in a series of new models enabled by its €3 billion (about $3.4 billion) Series D funding round last month, and that further reinforcement learning is likely to produce a more capable Mistral Large 4.1.