Mistral unveiled its Mistral Large 4 (ML4) model Tuesday, a 1-trillion-parameter open-weight system trained on 4,000 Nvidia Grace Blackwell GPUs over two months in European data centers. The company is positioning it as the strongest non-Chinese open offering, targeting enterprise and government buyers in cyber defense, coding, manufacturing and finance.
Open models—systems that can be modified and self-hosted—have become a battleground between Western and Chinese AI developers. Chinese open models have outpaced Western alternatives in capability and adoption. Mistral's bet is that enterprises and governments will favor a European model over closed U.S. systems or China's offerings.
The timing reflects Mistral's financial runway. The company raised 3 billion euros ($3.4 billion) in Series D funding in September, valuing it at 21 billion euros. That capital directly funds the GPU infrastructure for models like ML4—a costly bet on a market still consolidating around a handful of capability leaders.
ML4 will enter limited preview for developers, cybersecurity officials and state authorities this week, with a wider release later in the month. Mistral claims the model will rank among the top open-weight systems globally on aggregate benchmarks once weights are public. But the company acknowledges gaps remain: coding performance lags frontier models, suggesting continuous R&D spending will be necessary to maintain competitive positioning.
Guillaume Lample, Mistral co-founder and chief scientist, said the model's performance will improve as the company scales training following the fundraise. He highlighted cyber defense features as a key differentiator—enabling enterprises and governments to detect threats from attackers using jailbroken closed-model systems.
The open-source model category has gained strategic urgency. Earlier this year, Hugging Face deployed a Chinese GLM model to defend against a breach after encountering guardrails in leading U.S. closed systems. Separately, Reflection AI, an American startup backed by Nvidia, launched its own open model Monday, claiming parity with GLM 5.2—though GLM 5.3 surpassed it by August. The churn suggests no clear winner has emerged in open-weight AI, leaving room for a well-capitalized European entrant like Mistral to gain traction.
