Mistral launches Large 4, an AI model with 1 trillion parameters to compete with US and Chinese giants

French company Mistral announced its latest artificial intelligence system, Mistral Large 4 (also known as “Le Chonk”), on Tuesday, October 6. The model, which has around 1 trillion parameters, is already available for testing via the company’s API, with the weight files set to be released to the public on October 27.

With this new version, the company aims to position itself alongside the world’s leading models in corporate tasks such as programming, financial analysis, and digital security. Large 4 is described as Mistral’s largest and most powerful model to date, capable of processing both text and images and maintaining a context window of up to 1 million tokens.

The system’s architecture uses the Mixture of Experts technique, which activates only a portion of the parameters for each operation, allowing for a total size of 1.05 trillion parameters, of which 49 billion remain active during processing. This approach reduces computational costs and enables the model to run on customers’ own infrastructure.

Features and applications

Large 4 offers features for document querying, tool integration, and operation with autonomous agents. In DeepSWE 1.1 programming tests, the model achieved a 61.7% success rate. In computer vision, the company claims the system outperforms even some leading closed models in locating objects and visual elements.

Digital security is a key focus. Mistral had previously expressed interest in the sector by negotiating vulnerability detection solutions for European banks. The test version will be used in real-world scenarios with cybersecurity experts, selected partners, and public authorities, who will have access to a variant with fewer moderation restrictions and enhanced protection features.

Infrastructure and investment

Large 4 was trained from scratch on 3,800 Nvidia Grace Blackwell GPUs, distributed across the company’s European data centers. The announcement follows a €3 billion funding round, completed in September and led by Samsung Electronics, which raised Mistral’s valuation to over €21 billion.

The release of the model’s weights in October will enable organizations to run the system in their own environments, without relying on Mistral’s hosted service. Until then, the company will continue to conduct practical tests to refine the model’s performance and security.

Context and impact

The launch reinforces Mistral’s strategy of providing a European alternative to the large AI models developed in the United States and China. By offering a large-scale model that can be adapted and hosted locally, the company aims to attract corporate customers seeking control over their data and processes.

The initiative also reflects the growing investment in artificial intelligence in Europe, where governments and businesses are striving to balance technological innovation with strategic autonomy and the protection of sensitive information.