Key points
- Mistral says its new model rivals China's best open AI
- Its pitch leans on cybersecurity tests
- The weights are due Oct. 27
French AI company Mistral released a public preview of Mistral Large 4 on Oct. 6, saying it rivals the world's strongest open models and significantly outperforms those developed in the US or Europe. Nicknamed "le Chonk," the model has 1 trillion parameters, with 49 billion active at a time, according to Mistral's announcement.
Chief Executive Arthur Mensch highlighted cybersecurity at a conference in Abu Dhabi. "The model we're actually announcing today is actually above the Chinese models on certain aspects, including cyber," he said, Reuters reported. He didn't name the models or specify how they were compared.
Mistral plans to release the weights on Oct. 27, according to Reuters. During the preview, cybersecurity firms, vetted partners and state authorities are testing a version with reduced moderation and expanded cyber capabilities, the company said.
Mistral takes on cyber tasks Claude and GPT-6 refuse
On a test that asks a model to reproduce a real software vulnerability and then patch it, Mistral Large 4 scored 82%, the highest of any model, according to Mistral. Several closed models, including Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Astra, "score near zero on the same test because they refuse to perform the task," the company wrote. That gap reflects differences in what the models will agree to do, so it doesn't establish the same gap in technical ability.
Mistral said the model ranks among the top five on the Artificial Analysis Cyber Index, which it described as an independent evaluation. Most of the other scores in the announcement are Mistral's own.
Claude led Mistral's blind coding test
Mistral Large 4 ranked second among five models in a blind coding evaluation Mistral conducted with Surge AI. Professional reviewers gave it 3.74 out of 5, behind Claude Opus 5 at 4.22 but ahead of GLM-5.3 at 3.60 and Kimi K3 at 3.59.
The release also follows another Western push into open-weight AI. A day earlier, Nvidia-backed Reflection unveiled Beam, which it says matches Z.ai's GLM-5.2 on advanced reasoning while using far less compute. Mistral didn't include Beam in its published comparisons.
Mistral trained it on Nvidia chips in Europe
Mistral trained the model from scratch on 3,800 Nvidia (NVDA) Grace Blackwell GPUs in its own data centers in Europe, and it's serving the preview from the same hardware. Mistral said it will also offer a European deployment that it runs itself, under European law.
The company raised 3 billion euros at a 21 billion euro valuation in September, in a round led by Samsung. Mensch told the Abu Dhabi audience that Mistral's independence from the US and China is helping it grow in the Gulf and Asia-Pacific. "There's an enormous desire and an enormous need for alternative technology suppliers," he said.
