French artificial intelligence company Mistral has unveiled its latest flagship AI model, Mistral Large 4, positioning the system as a competitor to some of the most powerful Chinese open-weight models currently available.
The company announced the new model on Tuesday, describing it as its largest and most capable AI system to date. Mistral said the model, which is also known by the nickname “Le Chonk,” has been designed to perform across a wide range of demanding applications, including software development, cybersecurity, finance, legal work and image understanding.
According to the company, Mistral Large 4 contains approximately 1 trillion parameters, the learned numerical values that help an AI model process information and generate responses. Despite its enormous overall size, approximately 49 billion parameters are active during processing, a design intended to deliver substantial capabilities while making the model more efficient to operate.
One of the model’s most notable features is its reported 1-million-token context window. This allows the system to process extremely large amounts of information within a single request, potentially making it particularly useful for organizations working with lengthy documents, large codebases, extensive research material and other substantial datasets.
The company said developers can access a public preview of Mistral Large 4 through Mistral Studio, its development platform. Mistral plans to make the model fully public by the end of October.
Before the wider release, the company is testing the system with cybersecurity specialists, selected partners and government authorities. The testing programme is intended to assess the model’s capabilities and identify potential risks before broader deployment.
Mistral’s decision to emphasize the model’s open-weight nature is also significant. Open-weight AI models allow organizations to obtain the model’s learned parameters and deploy them on their own infrastructure, rather than relying exclusively on a provider’s cloud-based services.
This approach can give businesses and institutions greater control over their data, deployment environment and customization options. It can also be particularly attractive to governments and companies with strict requirements concerning data sovereignty, privacy and security.
Mistral said Large 4 was trained entirely from scratch using 3,800 Nvidia Grace Blackwell graphics processors located in the company’s own European data centers.
The company said its training data covered more than 160 languages, including every official language of the European Union. The broad multilingual training approach is designed to strengthen the model's ability to operate across different linguistic environments and support users throughout Europe and beyond.
The launch represents another major step in Mistral’s efforts to establish itself as a leading European competitor in the rapidly expanding global AI industry.
The company has increasingly positioned its technology as an alternative to AI systems developed by major US and Chinese technology companies. Its emphasis on open-weight models also distinguishes its strategy from providers that keep the underlying parameters of their most advanced systems private.
Mistral’s latest release comes shortly after the company secured major new investment. In September, the French AI firm raised €3 billion ($3.37 billion) in a Series D funding round led by Samsung Electronics.
The fundraising valued Mistral at more than €21 billion, underlining investor confidence in the company's prospects as demand for advanced AI systems continues to grow.
The latest model is being introduced at a time of intense competition among AI developers worldwide. Companies in the United States, Europe and China are racing to develop increasingly capable systems while simultaneously seeking to make them more efficient, affordable and customizable.
Mistral’s claim that Large 4 can compete with leading Chinese open-weight models places the French company directly within that international competition.
The model's combination of a trillion-parameter architecture, a million-token context window, multilingual capabilities and open-weight availability could make it particularly attractive to developers and institutions seeking to deploy advanced AI systems under their own control.
Cybersecurity is another major focus of the launch. Mistral said it is working with cybersecurity experts and other vetted organizations ahead of the model's broader release, reflecting growing industry concerns about the potential misuse of powerful AI systems.
As governments and companies increasingly incorporate AI into sensitive areas such as finance, law, software development and national security, the ability to deploy models securely and maintain control over sensitive information is becoming an important consideration.
With Mistral Large 4, the French company is therefore seeking not only to demonstrate technical progress but also to strengthen its position as a major European player in the global race to develop advanced artificial intelligence.
