The European race to reach the world's leading edge in artificial intelligence is gaining new momentum. The French company Mistral AI has unveiled its most powerful system yet, the Mistral Large 4 model. This innovation boasts over a trillion parameters and positions the company as a direct competitor to the most advanced American and Chinese models.

The French startup Mistral AI has introduced its new flagship model, Mistral Large 4, internally nicknamed "Le Chonk." This is the first major release from the company in approximately five months and represents one of the most ambitious European projects in the field of generative artificial intelligence.

The model contains approximately 1.05 trillion parameters, and thanks to its Mixture-of-Experts architecture, only about 49 billion of these are active when processing each token. This approach aims to combine high performance with more efficient use of computing resources.

Mistral Large 4 is also multimodal. It can work not only with text but also with image inputs, and according to publicly available technical information, it offers a context window of up to approximately one million tokens. This allows it to process very large documents, long conversations, or vast amounts of data in a single request.

Mistral aims to compete with the USA and China

Mistral has been striving to position itself as a European alternative to dominant American companies such as OpenAI, Google, or Anthropic, as well as rapidly growing Chinese laboratories.

According to Reuters, the company claims that Mistral Large 4 outperforms some competing open-source models, particularly in the area of cybersecurity. The company also highlights results in programming, financial tasks, geospatial analysis, and computer chip design.

Mistral is currently publishing its own test results, and a broader independent comparison of the new model will follow. It is through real-world performance that we will see how close the European model has come to the world's leading edge.

However, the significance of Large 4 extends beyond just benchmarks. Mistral is striving to promote the idea that Europe does not have to rely solely on technologies from the United States or China in the field of advanced AI.

The model was created using European infrastructure

Of particular interest is the way in which Mistral Large 4 was created. According to the company, the model was trained on its own infrastructure within European data centers.

Technical documentation indicates the use of approximately 3,800 NVIDIA Grace Blackwell GPUs. Pierre Stock, Vice President of Science at Mistral, told TechCrunch that the company managed to train the model with significantly fewer GPUs than some Chinese or American competitors reportedly use.

For Mistral, control over technological infrastructure is a key part of its strategy. The company wants to offer models that can be used not only by technology companies but also by European governments, banks, industrial enterprises, and other institutions working with sensitive data.

The model's weights will be publicly available

Mistral Large 4 is now available in a public preview mode through the company's interface. A significant step is planned for the end of October.

Mistral plans to release the model's weights on October 27, which will allow it to be run outside of the company's own infrastructure. Open weights are important, for example, for companies or government institutions that want to operate artificial intelligence on their own servers and do not want to send sensitive data to external providers.

However, it will undergo further security testing before its release. According to Reuters, Mistral will initially provide broader access to cybersecurity experts and collaborating institutions.

The company is responding to one of the main problems in current AI development. Powerful models can help identify security vulnerabilities and defend information systems, but they can also potentially be used for attacks.

European AI Secures Billions of Euros

The launch of Large 4 comes just a few weeks after Mistral announced that it had secured three billion euros in new funding. The investment round valued the company at over 21 billion euros.

Notable supporters of Mistral include, for example, the Dutch manufacturer of equipment for the chip industry, ASML, and the South Korean company Samsung.

The French company is also investing billions of euros in its own infrastructure and data centers. Its goal is not only to create individual models but also to gradually build a European technological ecosystem capable of competing with the largest global players.

Mistral Large 4 is therefore interesting not only for its more than one trillion parameters. It is also a test of whether Europe can create its own cutting-edge artificial intelligence while maintaining control over the infrastructure on which this technology will increasingly rely in the future.

Jaroslav Holý