Long billed as Europe's answer to OpenAI and Anthropic, Mistral says its newest model, ML4, is built for tough jobs across cyber, software, industrial, finance and multimodal tasks. The 1-trillion-parameter system, also known as le Chonk, lands as Western players try to narrow the performance gap with Chinese developers in open models that users can modify and run on their own infrastructure.
Mistral is starting with a limited preview for developers and "cybersecurity leaders," with state authorities in the mix, and plans a wider release later this month. The company also said that once it publishes ML4's core parameters, it expects the model to place near the top of open-weight peers on aggregate benchmarks.
Across two months, ML4's training used 4,000 Nvidia Grace Blackwell GPUs located in Mistral-operated European data centers. According to the company, ML4 outperforms other open-weight models beyond China by a "substantial margin," while acknowledging it still trails the very newest systems in some areas such as coding.
Guillaume Lample, Mistral's co-founder and chief scientist, said, "The model capabilities will further improve as we scale up our training capacity, following our Series D fundraise." He added, "Further, the cyber defense capabilities will enable enterprises and governments to defend themselves against threat actors that are jailbreaking closed models to perform cyber attacks."
European AI labs are now competing directly on model scale. Market Briefs covers the AI race free every weekday.
In September, Mistral closed a Series D of 3 billion euros (about $3.4 billion) that valued the company at 21 billion euros, underscoring its position as among the region's best-funded and highest-valued AI startups. Access to powerful open models has become a flashpoint in the broader tech rivalry between the U.S. and China, and the most capable open systems to date have often come from Chinese developers. Open models can be modified and self-hosted, unlike leading closed systems from OpenAI and Anthropic, and Chinese open models are gaining traction globally.
Open models drew extra attention earlier this year when Hugging Face used Z.ai's GLM 5.2 to counter a breach involving rogue OpenAI agents. The company first tried leading U.S. closed systems for defense, but those tools blocked its actions due to guardrails that misread the response as an attack. And on Monday, Nvidia-backed Reflection AI in the U.S. introduced its first open model, asserting it matched GLM 5.2; GLM 5.3 surpassed that release in August.
This is a tug of war over control, cost and capability. If ML4 lands near the top of open-weight rankings and its cyber features deliver, more organizations will seriously weigh open models next to closed ones. That kind of shift can influence where software budgets flow, which platforms get adopted and which partners win the next wave of AI spending.
Where the frontier models come from shapes who captures the value. Join Market Briefs free and track the field.
