French AI startup Mistral declared on Tuesday its biggest open-weight model, dubbed “Chonk,” claiming enhanced overall performance and security capabilities for customers intending to maintain their data private rather than of depending on closed models from the likes of Anthropic and OpenAI.
The new model lets customers download to their own servers and customise the parameters, or “weights,” that determine how it computes.
It is Mistral’s first major product release since it raised 3-billion euros ($3.4 billion) last month in Europe’s biggest technology fundraising ever.
The company is making a bet that companies and governments will an increasingly want total control over their AI deployments, instead of the using proprietary models that need delivering over data to third parties and can be much more costly.
“Enterprises across financial services, production and the public sector can govern the intelligence, data, compute and operations underpinning their essential infrastructure … Without exposing their most precious knowledge to anyone outside their own walls,” the company mention in a statement.
“That one took some groundwork,” Mistral’s chief executive Arthur Mensch published on X.
“Trained and served on our very own compute, and RL shows no sign of saturation,” he wrote, regarding reinforcement learning to further expand AI models and to the possibility that adding similarly data starts to produce diminishing performance returns.
Open-weight models are also touted as more stable, a growing concern after numerous incidents of AI agents going rogue by hacking databases or otherwise doing things they were not supposed to do.
“The cyber defense supposed will permit enterprises and governments to defend themselves against against actors that are jailbreaking closed models to carry out cyberattacks,” Mistral’s chief scientist Guillaume Lample said in the statement.
Data ‘under their control’
Mistral, considered Europe’s main AI developer, was valued at over 21 billion euros in last month’s investment around—far below the valuations for U.S. Giants like OpenAI or the AI operations of Google, Meta and SpaceX.
Anthropic is reportedly getting ready for an initial public providing within the coming weeks that could see it valued at $2 trillion.
But their so-called closed models that presently dominate the market are dealing with developing competition from open-weight models, largely from Chinese developers along with DeepSeek and Moonshot.
These AI versions may be fine-tuned via clients for specific requirement on in-house infrastructure and used at a fraction of the cost.
Mistral’s new model also comes as a U.S. Open-weight rival, Reflection AI, declared its own broadly expected Beam model on Monday, also promising cheaper operating costs.
Mistral said Chonk’s scores on overall performance benchmarks so far make it one of the high-quality open-weight models available, and “by a long way the best in Europe and the U.S.”
It said the model, which remains being subtle, was constructed on 4,000 Nvidia chips over 2-months at its own data centers in Europe.
Customers can either download the model to their own servers or access of via Mistral’s infrastructure, which includes in “our European sovereign region where data remains under EU jurisdiction to make sure data stays below their control.”
Mistral said the model was trained on more than 160 languages and can answer prompts and questions in each one natively.
