Mistral has launched a public preview of Mistral Large 4, a one-trillion-parameter multimodal model code-named "Le Chonk," VentureBeat reports. The Paris lab says the model pushes it back toward the global open-weight frontier, with a focus on coding, cybersecurity, finance, manufacturing and visual grounding. WIRED adds that Mistral presents Le Chonk as the most capable open-weight model developed outside China and "very, very close" to some proprietary systems. The nickname deliberately nods to the viral "Le Chaton Fat" meme from June, a fictional fat-cat model the community invented; VentureBeat notes CEO Arthur Mensch and investor Marc Andreessen amplified the joke, and co-founder Guillaume Lample suggested ML4 can be viewed as an initial version of that imagined model, with still larger ones to come.

Mistral is pitching open weights specifically for security teams, VentureBeat reports. Its argument: enterprises cannot depend entirely on closed providers whose safety systems may refuse dual-use but legitimate defensive requests, so an open-weight model gives security teams greater control over code scanning, defensive testing and other high-volume security workflows without depending on a provider's changing moderation policy. Lample told WIRED that the stakes extend beyond Europe: "If you use a closed model, there is no guarantee it will still be there tomorrow." WIRED also reports the Trump administration temporarily restricted distribution of OpenAI and Anthropic models in June over cyberattack concerns, and that the White House reportedly asked American labs to withhold unreleased models even from the UK's AI Safety Institute — reminders that access to frontier AI can be unilaterally revoked, which WIRED says has opened the door for a Europe-based open-weight lab.

ML4 arrives on a hot streak. VentureBeat and WIRED both report that in September Mistral raised about €3 billion (roughly $3.3 billion) at a post-money valuation above €21 billion — about $24 billion per Reuters — the largest equity raise by a European tech company. VentureBeat says Mistral now supports more than 125 enterprises including Airbus, ASML and HSBC, and that its stack has grown to include developer products, customization services, inference infrastructure and Mistral Compute. According to VentureBeat, Lample has scaled the science team from three researchers to roughly 300, and Lample told VentureBeat that customers increasingly need deployment, infrastructure and engineering support around complex AI workflows — Mistral is betting weights commoditize while the surrounding enterprise business carries the value.

VentureBeat reports the weights are expected under a custom Mistral license on October 27, after roughly three weeks of testing with developers, cybersecurity leaders and government authorities; the preview period will also let Mistral continue reinforcement learning and tune the final checkpoint. WIRED confirms a final version is due by the end of the month. Meanwhile, the release lands amid a widening Western open-weight push: TechCrunch and Fortune report Reflection AI debuted Beam, a 501-billion-parameter open model, on October 5, meaning ML4 already has a direct Western rival's numbers in its own benchmark charts. The decisive test, as VentureBeat frames it, is whether ML4's final checkpoint holds up once independent evaluators can run it — after October 27, the nickname matters less than what developers can reproduce on their own hardware.