← Today’s paper

From the paper of Wednesday, October 7, 2026

Mistral releases a trillion-parameter model whose open weights arrive in three weeks

No. 4·422 words·2 min

Mistral AI released Mistral Large 4 on Tuesday, a one-trillion-parameter multimodal model that the French lab says is the most capable open-weight model developed outside China and "very, very close" to the proprietary models from OpenAI and Anthropic. The weights that would make it open are not out yet.

Nicknamed Le Chonk for its size, the model is available in preview, and for now can be reached only through a public guardrail endpoint. Mistral plans to publish the weights in about three weeks, after safety testing is complete, Pierre Stock, the company's vice president of science, told TechCrunch. Wired reported the model as freely available, usable and customizable by anyone. Mistral's benchmark results are still pending.

"In the meantime, we'll work with trusted partners and governments to make sure that the open source weights can be used to defend, but not to [perform] malicious attacks," Stock said.

Mistral says it optimized the model for coding, cybersecurity, finance, manufacturing, and electrical engineering, along with chip design — a specialty of two backers, ASML, which led its Series C round, and Samsung, which led its Series D last month at a valuation of €21 billion, about $24.4 billion. Wired reported that the September round raised $3.3 billion at a $24 billion valuation, which it described as the largest ever raised by a European tech company.

Mistral says Le Chonk was trained from scratch on its own compute, using 4,000 Nvidia GPUs, which Stock said is two to three times fewer than its Chinese competitors and significantly fewer than the closed-source labs. The claim is pointed: the US government has accused Chinese labs of abusing distillation — training a smaller model on the outputs of a larger one — to close the gap with OpenAI and Anthropic.

"Sometimes, people like to [make a big deal] over the US, versus Europe, versus China. But what really matters is to own the model—even for US companies," cofounder and chief scientist Guillaume Lample told Wired. "If you use a closed model, there is no guarantee it will still be there tomorrow."

In June, the Trump administration placed temporary restrictions on the distribution of models from OpenAI and Anthropic, citing concerns they could be abused to launch sophisticated cyberattacks. Wired reported that since then, US-made models have broken free of their constraints and attacked companies and some foreign government institutions, leading to debate over how releases should be regulated. The White House has reportedly asked the American labs to withhold unreleased models even from the UK's AI Safety Institute.

Discuss on the board