ARTIFICIAL INTELLIGENCE
Mistral Releases Large 4, a 1-Trillion-Parameter Open-Weights Model Dubbed 'Le Chonk'
French AI lab Mistral has unveiled Large 4, a massive multimodal model it says leads open-weights offerings from the US and Europe and beats closed rivals on visual grounding. Open weights are due at the end of October.
Image: Mistral · Uploaded by IntraGoals — usage rights confirmed
Mistral AI has announced Large 4, a new flagship language model the company is informally calling "Le Chonk" for its enormous scale. According to the company, the model has 1 trillion total parameters with 49 billion active at any given time, and it is natively multimodal, meaning it can process and reason across different types of input, such as text and images, without relying on separate bolt-on systems.
Mistral says Large 4 is the best open-weights model available from either the United States or Europe when measured against aggregated benchmarks. The company also claims state-of-the-art performance on workloads it describes as critical, including cyber defense, manufacturing and finance, and says the model surpasses closed frontier systems on visual grounding, the ability to accurately connect language to specific objects or regions within an image.
The company is positioning Large 4 as a distinctly European product, saying it was "forged in Europe end-to-end" and can be deployed from Europe using Mistral's own Mistral Cloud infrastructure. That framing comes as European governments and businesses have increasingly sought AI providers that offer alternatives to the dominant American cloud and model providers.
Large 4 is available immediately through Mistral's API, and the company says it is working privately with cybersecurity partners ahead of a broader release. Mistral plans to make the model's open weights available by the end of October, which would allow developers and researchers to download, inspect and run the model independently rather than accessing it only through Mistral's hosted service.
The announcement was made via Mistral's official account on X. Specific benchmark figures, independent verification of the performance claims, and further technical details about the model's architecture were not included in the company's statement and have not yet been independently confirmed.