Mistral Launches Large 4 AI Model, Challenging Chinese and US Rivals
Mistral Large 4 Arrives in Public Preview
Mistral has launched Mistral Large 4, its latest flagship AI model and the company's biggest release since May. The model, nicknamed “Le Chonk,” is being offered initially as a public preview through Mistral's API, with its weights scheduled for release on October 27.
Mistral is positioning Large 4 as a European alternative to both US-built proprietary models and Chinese open-weight systems. CEO Arthur Mensch said the model outperforms Chinese models in some areas, particularly cybersecurity, although he did not identify the specific models or benchmarks behind that claim.
One Trillion Parameters, 49 Billion Active
Mistral Large 4 is a natively multimodal Mixture-of-Experts model with around 1 trillion total parameters and 49 billion active parameters. It is designed for general-purpose use across text and visual workloads, including coding, reasoning and agentic tasks.
Mistral says the model is particularly strong on enterprise workloads such as cybersecurity, finance and law, while also targeting manufacturing and visual understanding. The company trained it from scratch using 3,800 NVIDIA Grace Blackwell GPUs in its European data centers.
Cybersecurity Is a Major Focus
Cybersecurity is one of the most prominent areas in Mistral's pitch for Large 4. The company says the model ranks among the world's strongest AI systems for cybersecurity and leads open-weight models developed outside China on the Artificial Analysis Cyber Index.
Mistral reports an 82% score on a test requiring models to reproduce a real software vulnerability and then patch it, while also reporting that Large 4 solved 93% of challenges in Cybench. These are benchmark results cited by Mistral and should not be interpreted as proof that the model is universally superior at cybersecurity.
The company is allowing cybersecurity experts, government authorities and selected partners to test a version with reduced safety restrictions and expanded cyber capabilities before the model's weights are released. The testing is intended to identify risks associated with giving the model stronger offensive-security capabilities.
Open Weights Strengthen Mistral's European Pitch
The planned open-weight release is central to Mistral's strategy. Rather than requiring organisations to depend entirely on a proprietary API, the company intends to give customers access to the model weights so they can deploy the system under their own infrastructure and policies.
Mistral says this is particularly important for cybersecurity and other sensitive workloads where organisations may need greater control over data, model behaviour and availability. The company also plans a European deployment operated independently under European law.
Large 4 was trained on multilingual data spanning more than 160 languages, including all official languages of the European Union. Mistral is therefore combining model performance with European infrastructure and deployment control as it seeks to differentiate itself from US and Chinese competitors.
Mistral Targets the Open-Weight AI Race
The launch comes after Mistral went roughly five months without releasing a major new model. During that period, the company expanded access to third-party models and raised €3 billion at a €21 billion valuation in September.
Large 4 gives Mistral a new vehicle for competing in the increasingly important open-weight segment. Independent analysis from Artificial Analysis puts the model at 38 on its Intelligence Index, while its cyber score is particularly strong. Those results vary substantially by workload, meaning Large 4's strongest performance areas should not be treated as a universal lead over every US or Chinese model.
For now, the model remains in preview. The October 27 weight release will be the next major test of Mistral's open-weight strategy and will allow developers to evaluate the model outside the company's hosted environment.
