Mistral AI Announces New AI Model with 1 Trillion Parameters

The French company Mistral AI has announced a public preview of its AI model, Mistral Large 4, which boasts 1.05 trillion parameters and open weights. The model is positioned as a competitor to leading American and Chinese neural networks, focusing on tasks related to programming, cybersecurity, and the corporate sector. Mistral Large 4, also known as Le Chonk, features 49 billion active parameters and a separate visual encoder with 1.6 billion parameters, supporting a context of up to 1 million tokens. The company claims that this model is among the most efficient open-weight solutions and narrows the gap with closed systems, outperforming some Chinese solutions in the field of cybersecurity. However, Mistral has not provided independent results to substantiate its claims.

The training of Large 4 was conducted on 4,000 Nvidia Grace Blackwell GPUs over approximately two months in the company's European data centers. The model covers a wide range of tasks, including programming, finance, geospatial analysis, and engineering, and is developed with a focus on agent scenarios requiring the processing of large data volumes. One of the main advantages of Large 4 is the availability of open weights, allowing developers to run and adapt the model without transferring data to third-party providers. The model is currently available in public preview mode, with the full release of weights scheduled for October 27. Until then, Mistral is offering early access to developers and organizations working in the field of cybersecurity.

The release of Mistral Large 4 comes amid growing competition in the open-weight model segment, where Chinese companies like Alibaba and Z.ai are significantly strengthening their positions. Mistral aims to occupy an intermediate position, offering models with the functionality of advanced systems while providing users with control over parameters and infrastructure. Recently, the company secured €3 billion in investments, valuing itself at over $24 billion.