French startup Mistral unveiled its new Large 4 artificial intelligence model on October 6 in Abu Dhabi, which Pierre Stock, the company's vice president of research, said is capable of designing computer chips. The company promises to make the model publicly available on October 27, according to infohub.kz.
Stock said Mistral intends to deepen cooperation with its investors, including ASML and Samsung. He declined to comment on whether the company is working with Samsung on chip development.
According to Mistral, Large 4 narrows the gap with the most advanced AI models in programming, finance, geospatial analysis, industrial manufacturing, and product design.
CEO Arthur Mensch said at the presentation that the model outperforms some Chinese models.
"The model we are presenting today surpasses Chinese models in some parameters, including in cybersecurity. So the claim that Europe cannot compete is not true," he said.
Mensch did not name the specific models or benchmark results.
Large 4 was trained on Mistral's own infrastructure. The model will be released with open weights: users will be able to download it and run it on their own hardware.
Before the public release, cybersecurity experts and government agencies will receive a version with fewer safety restrictions to test its capabilities.
Stock said that during testing, the model attempted to escape the test environment. According to him, the company expected such behavior and was able to prevent it. Similar incidents were recorded during testing of advanced models from OpenAI and Anthropic.
Mistral has been in existence for three years and is considered one of Europe's main competitors to American and Chinese AI developers. In its latest funding round, the startup raised €3 billion — about $3.4 billion. Large 4 is its first major new model in about five months.
Earlier, Kursiv wrote about how the Danes launched their own European ChatGPT that does not transfer data to the US.