Mistral on Tuesday unveiled a new AI model that it said ranks among the world’s best open models, as Western companies race to close the capability gap with Chinese developers in the market.
Mistral Large 4 (ML4), also known as le Chonk, is a trillion-parameter model that the French company says is particularly effective for tasks related to cyber, coding, manufacturing, finance, and multimodal.
Mistral, long considered Europe’s only answer to OpenAI and Anthropic, raised a €3 billion ($3.4 billion) Series D in September at a valuation of €21 billion. This round solidifies the company’s position as one of the best-funded and most valuable AI startups in the region.
Access to so-called open AI models, the most capable of which are made in China, has become a key flashpoint in the escalating competition for technological superiority between the United States and China.
In contrast to the dominant closed systems created by OpenAI and Anthropic, open AI models can be modified and self-hosted. China’s open model is increasingly being adopted around the world.
Mistral Large 4 is being previewed for developers and “cybersecurity leaders” with state officials, the company said, before its wide release later this month.
The rise of the open model
ML4 was trained with 4,000 Nvidia Grace Blackwell GPUs were deployed at Mistral’s own data centers in Europe over a two-month period.
Mistral said in a statement that once the core parameters are made public, ML4 will rank among the top free-weight models in the world for overall benchmark performance. The company added that the ML4 is the most powerful open-class model developed outside China by a “significant margin.”
Newer models still lag behind the state of the art in areas such as coding.
“The capabilities of our model will continue to improve as we expand our training capabilities with our Series D funding,” said Guillaume Lampre, co-founder and chief scientist at Mistral.
“Furthermore, cyber defense capabilities will enable businesses and governments to protect themselves from attackers who jailbreak closed models and carry out cyber attacks,” he added.
Open models were thrust into the spotlight earlier this year when it was revealed that startup Hugging Face used China-based Z.ai’s GLM 5.2 to prevent breaches from rogue OpenAI agents. Hug Face initially turned to the United States’ primary closed systems to protect itself, but was stopped by Guardrails, which misinterpreted his actions as an attack.
Reflection AI, an American startup backed by Nvidia, on Monday announced its first open model, which it claims will match the performance of GLM 5.2, which was overtaken by GLM 5.3 in August.
