Mistral Large 4 Preview Released: Europe's First Trillion-Parameter Open-Source Model to Release Weights at Month's End

On October 6, 2026, France's Mistral AI released the Mistral Large 4 preview (nicknamed Le Chonk), with 1.05 trillion total parameters and 52B active parameters. It was trained on 3,800 Grace Blackwell GPUs in Mistral's own European data center, its API is already live, and its weights are expected to be open-sourced on October 27.

The Facts

Mistral Large 4 is a native multimodal model that supports input in more than 160 languages and has a context length of up to 1M. It scored 61.7% on the DeepSWE 1.1 benchmark, 28.3% on Terminal-Bench 4.0, and 59.4% on SWE-Atlas-QnA. The model ranks among the global top five in the Artificial Analysis Cyber Index and is the leading non-Chinese model among open-weight models. It scored 82% in vulnerability reproduction and remediation tests and solved 93% of challenges in the Cybench test.

Its training data includes a large amount of multilingual content, covering all official languages of the European Union. The model is independently deployed in European data centers, does not rely on other digital service providers, and is subject to European law. During the preview stage, it has conducted red-team testing with cybersecurity leaders, partners, and national agencies, giving it reduced moderation and expanded cyber capabilities.

Mechanism Breakdown

Le Chonk uses a MoE architecture, with 1.05 trillion total parameters but only 52B active parameters per inference; this design controls compute costs while maintaining high performance. The model was trained from scratch in Mistral's own European data centers, and the infrastructure is fully consistent with its inference service, reducing external dependence.

In cybersecurity, the model can perform vulnerability reproduction and remediation, tasks that closed-source models often refuse due to safety filters. Mistral emphasizes that such capabilities have practical value for defenders analyzing malware, prioritizing vulnerabilities, and writing detection rules.

Industry Impact

The model will become Europe's first trillion-parameter open-weight model, filling a previous gap between the United States and Europe in open models of this scale. API pricing is $1.36 per million input tokens and $4.18 per million output tokens, and it is now available on Mistral Studio.

After the weights are open-sourced, enterprises can deploy them in private clouds or on-premises environments to meet requirements for data sovereignty and auditability, especially in vertical sectors such as finance, law, and cybersecurity.

Strategic Assessment

[Analysis] Europe's breakthrough in open-weight models may change the U.S.-dominated global AI competition landscape, but the model's actual impact still depends on community adoption and benchmark performance after it is open-sourced at the end of the month. Mistral positions this model as the foundation for subsequent specialized models, and its open capabilities in sensitive areas such as cybersecurity may prompt closed-source vendors to reassess their safety filtering strategies.