The race between open and closed AI continues, and Europe is still in the middle of it. On Tuesday, French AI institute Mistral AI released Mistral Large 4, a new large multimodal model that aims to leapfrog its American and Chinese rivals, following what French President Emmanuel Macron has described as the “third way of AI.”
As the gap widens between unpluggable closed models and open models, often made in China, Mistral is positioning the ML4 as an alternative to both. Nicknamed Le Chonk for its trillion parameters, ML4 is no small thing. However, it is not yet an open weight model. Although currently only accessible via a public guardrail endpoint, Mistral plans to have the weights available in just three weeks after safety testing is complete.
“In the meantime, we will work with trusted partners and governments to ensure that the weight of open source can be used to defend, but not to (perform) malicious attacks,” Pierre Stock, vice president of science at Mistral, told TechCrunch.
Security concerns have increased in recent months, especially among businesses and institutions that are Mistral’s primary audience. At the same time, the zero-gravity model is easier to audit, Stock said.
Another important behind-the-scenes aspect is that ML4 was trained entirely on Mistral compute. It uses just 4,000 NVIDIA GPUs, “which is two to three times less than our Chinese competitors and significantly less than our closed source competitors,” Stock said.
Benchmark results are not yet available, but Mistral expects the ML4 to be best-in-class among free weight models, especially outside of China, but Stock said there’s more to it than that. Thanks to intensive training, they may even outperform closed models in specific areas that are important to customers and where multimodal capabilities can add value.
Stock said ML4’s optimized use cases include not only cybersecurity and finance, but also the core chip designs of Mistral’s two main backers: Dutch giant ASML, which led a Series C, and Samsung, which led a Series D last month at a valuation of 21 billion euros (about $24.39 billion).
At the time, the company tried to communicate that its decision to host a Chinese model was more than just a transformation into an inference provider. With Le Chonk in the mix, the company believes it should still be considered a frontier lab.
If you make a purchase through links in our articles, we may earn a small commission. This does not affect editorial independence.
