Mistral, the French artificial intelligence startup, on Tuesday introduced its new model and ended an interval of about five months between its main releases. The company claims the new model performs better than many competing models.
Mistral was founded three years ago and is widely regarded as the largest European rival to American and Chinese AI companies. However, it had not released a new model since May. In August it also began offering third‑party models to customers, the first of which was GLM 5.3 from Z.DotAI based in Beijing.
“The narrative of Europe’s incapacity is not true”
Arthur Mensch, CEO of Mistral, said at the unveiling in Abu Dhabi that the introduced model performs better than Chinese models in some aspects, including cybersecurity. He emphasized that the narrative that Europe cannot compete is not true. Mensch did not name the Chinese models he meant or the metrics for judging the contest, so this claim has not yet been independently verified.
The model, officially named “Mistral Grand 4” and internally known as “Lou Chang,” will be available to the public on October 27. Mistral said that before the public release it will allow cybersecurity experts and government officials access to a version with fewer safety restrictions to examine its capabilities.
Pierre Stock, Mistral’s chief scientific officer, said the model is among the most powerful of the available “open‑weight” models. In this approach, users can download the model from platforms and run it on their own devices. Many Chinese models follow this approach, while leading American models are typically closed and only accessible through their developers’ infrastructure.