By Roman Aglaganov
On 6 October, Paris-based Mistral AI opened a public preview of Mistral Large 4. Officially ML4, unofficially, in Mistral’s own words, “le Chonk”. It is the company’s largest model so far, and the one it hopes will put Europe back at the top table of open AI.
What it is
Big, but not all at once: ML4 has about a trillion parameters, the numbers a model learns, but only 52 billion are active at a time, which keeps it cheaper to run than its size suggests. It is natively multimodal, and trained on data in more than 160 languages, including every official EU language.
Made in Europe: Mistral trained it from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its own European data centres, and serves the preview from the same machines, including a European deployment it runs end to end under European law.
Open, soon: the preview is available through Mistral’s API, at $1.36 per million input tokens and $4.18 per million output tokens. Mistral says the weights will be released by the end of October.
The cybersecurity pitch
Mistral leans hard on security. It says ML4 ranks in the top five worldwide on the Artificial Analysis Cyber Index and solves 93% of Cybench, a set of 40 security-competition challenges. It also points out that some leading closed models score near zero on one test because they refuse to reproduce a vulnerability, which defenders often need to do to prove a flaw is real.
Before releasing the weights, Mistral says it is red-teaming the model with cybersecurity companies, vetted partners and state authorities, who get a version with reduced moderation. That is the open-weight trade-off in one sentence: a model defenders can run themselves, under their own rules, is also one attackers could eventually run under theirs.
Caveat: the benchmark figures come from Mistral and are preliminary. Independent tests will tell us more once the weights are out.
