The new AI model, nicknamed ‘Le Chonk’, brings advanced coding, multimodal capabilities and cybersecurity applications, while giving organisations greater control over deployment.
PARIS, October 30, 2026: French artificial intelligence company Mistral AI has released the open weights of its latest flagship model, Mistral Large 4, marking a significant development in the rapidly evolving AI industry.
Nicknamed “Le Chonk”, the model was initially introduced in a public preview on October 6, 2026. Its weights became publicly available on October 29, allowing developers and organisations to download and deploy the model on their own infrastructure.
The release is particularly notable because it gives organisations greater flexibility to run a highly capable AI model without relying entirely on a third-party hosted service.
A 1-Trillion-Parameter AI Model
According to Mistral’s technical documentation, Mistral Large 4 contains approximately 1.05 trillion total parameters, with 52 billion active parameters.
The model uses a Mixture-of-Experts (MoE) architecture, which activates selected parts of the network for different tasks rather than using the entire model for every operation.
This approach is designed to support large-scale AI capabilities while improving computational efficiency.
Mistral Large 4 also features a context window of up to one million tokens, enabling it to process extensive amounts of information within a single interaction.
The model is natively multimodal, meaning it can work with different types of information, including text and images.
Advanced Coding and AI Agent Capabilities
One of the major areas of focus for Mistral Large 4 is software development.
The company has highlighted its performance in coding, repository analysis and complex terminal-based workflows.
According to Mistral’s published results, the model achieved:
61.7% on DeepSWE v1.1.
59.4% on SWE-Atlas-QnA.
28.3% on Terminal-Bench 4.0 in the initial preview evaluation.
The company also reported improvements following additional reinforcement learning training before the open-weight release.
These capabilities could make the model useful for developers working on software engineering, debugging and automated coding tasks.
Beyond coding, Mistral Large 4 is designed to support AI agents that can gather information, interact with tools and complete multi-step workflows.
A Strong Focus on Cybersecurity
Cybersecurity is another major area highlighted in Mistral’s announcement.
The company says Mistral Large 4 can assist with identifying software vulnerabilities, analysing security issues and developing fixes.
According to Mistral, the model achieved an 82% score on a test involving the reproduction and patching of a real vulnerability in open-source software.
It also reported a 93% success rate on Cybench, a benchmark consisting of cybersecurity challenges.
However, these figures should be understood in the context of the company’s published evaluations. Independent testing across broader real-world conditions will be important to establish how consistently the model performs.
The model’s advanced cybersecurity capabilities also raise questions about responsible deployment and the potential misuse of powerful AI systems.
Why Open Weights Matter
The release of Mistral Large 4’s weights is arguably its most important development.
In traditional hosted AI services, users generally access models through a company’s website or API. The provider controls the underlying infrastructure and model deployment.
Open-weight models offer a different approach.
They allow developers and organisations to obtain the trained model weights and deploy the system within their own computing environments, subject to the applicable licence terms.
For businesses handling confidential information, this can offer greater control over data processing, infrastructure and customisation.
It could also reduce dependence on individual AI service providers.
However, running a model of this scale independently requires substantial computing infrastructure, technical expertise and operational resources.
Europe’s Growing Ambitions in Artificial Intelligence
Mistral Large 4 also reflects Europe’s broader ambition to strengthen its position in the global AI industry.
According to the company, the model was trained using 3,800 NVIDIA Grace Blackwell GPUs in its European data centres.
Mistral has positioned the release around AI sovereignty, giving organisations greater control over where and how their AI systems operate.
This is particularly relevant for European businesses and public institutions seeking alternatives to dependence on a limited number of major international technology providers.
The company also says the model’s training included more than 160 languages, reflecting its focus on multilingual AI capabilities.
What This Means for the Future of AI
The release of Mistral Large 4 highlights an important shift in the AI industry: competition is no longer limited to building increasingly powerful chatbots.
Companies are now competing to develop AI systems capable of coding, analysing complex information, supporting scientific research and performing multi-step professional tasks.
At the same time, the growing availability of open-weight models is creating new opportunities for developers, enterprises and researchers.
While significant computing requirements remain a barrier to widespread independent deployment, Mistral Large 4 demonstrates how advanced AI capabilities are becoming available through more flexible deployment models.
The next challenge will be determining how these systems perform in real-world environments, how safely they can be deployed and whether they can compete with leading proprietary AI services.
Sources: Mistral AI official announcement and technical documentation; SiliconANGLE’s October 29, 2026 report.
