Mistral released a new artificial intelligence model on Tuesday, stating that the model is among the best open models in the world. Currently, Western companies are competing to narrow the gap in capabilities with Chinese developers in this market.

This French company stated that Mistral Large 4 ( ML4 ), also known as “ le Chonk ”, is a model with 1 trillion parameters that is particularly adept at tasks related to cybersecurity, programming, manufacturing, finance, and multimodal applications.

For a long time, Mistral has been regarded as Europe's only powerful response to OpenAI and Anthropic. The company completed a $3 billion (3.4 billion euro) Series D financing in September, with a valuation of 21 billion euros. This round of financing further solidified its position as one of the best-funded and most highly valued AI startups in Europe.

As the competition between the United States and China for technological dominance intensifies, obtaining so-called open-source AI models—most of which are from China—has become a key focus.

Different from the leading closed-source systems launched by OpenAI and Anthropic, the open AI model can be modified and self-hosted. China's open models are being adopted more and more widely around the world.

Mistral indicates that Mistral Large is currently available in preview version to developers, "cybersecurity officials," and government agencies, and will be more widely released later this month.

The Rise of Open Models

Mistral indicates that ML4 used 4,000 NVIDIA Grace Blackwell GPU deployed in its own European data center to train for two months.

In its statement, the company claimed that after the release of its core parameters, ML4 will rank among the world's top open-weight models based on comprehensive benchmark tests. The company also stated that ML4 is the strongest open-weight model developed outside of China, with "significant advantages."

However, this new model still lags behind the most advanced levels in fields such as programming.

Mistral, co-founder and chief scientist, Guillaume Lample, stated: "As we expand our training capabilities after the Series D financing, the model's performance will further improve."

He also stated, "In addition, cyber defense capabilities will enable enterprises and governments to withstand threat actors who are carrying out cyber attacks through bypassing closed-source models."

Earlier this year, open-source models received widespread attention. At that time, there were reports that the startup Hugging Face turned to use GLM version 5.2 from Z.ai, which is headquartered in China, when dealing with intrusions from out-of-control OpenAI agents. Hugging Face initially tried to defend itself using leading closed-source systems from the United States, but the security mechanisms of these systems mistakenly identified their actions as attacks, thereby preventing the relevant operations from proceeding.

A US startup supported by NVIDIA, Reflection AI, released its first open-source model on Monday, claiming that the capabilities of this model are comparable to those of GLM 5.2; meanwhile, GLM 5.2 was surpassed by GLM 5.3 in August.