Mistral AI has opened a public preview of Mistral Large 4, a new flagship model the company itself has nicknamed "Le Chonk". It launched on 6 October and is, by parameter count, the largest open-weight model Mistral has built: roughly 1.05 trillion parameters in total, using a mixture-of-experts design that activates only around 49 billion of them for any single token. A 1.6-billion-parameter vision encoder makes it multimodal, and it ships with a one-million-token context window.
The name has an unusual origin. Before launch, people on Reddit and X joked about a fictional Mistral model called "le chaton fat" ("the fat kitten"). Mistral's chief executive, Arthur Mensch, played along online, and the company then adopted a version of the joke as the model's own nickname at launch.
For now, Mistral Large 4 is reachable only through the company's own API; the full open weights are due by the end of October, so nobody can self-host it yet. That is a notable gap for a model marketed on its open-weight status. Training reportedly used several thousand Nvidia Grace Blackwell GPUs, underlining how much compute a trillion-parameter model demands even with sparse activation.
Mistral is one of the few major labs still releasing frontier-scale models with eventual open weights, rather than closing them entirely. A preview-now, weights-later release lets the company claim the scale milestone immediately while it finishes the safety and infrastructure work needed before handing the model to anyone to run themselves.
Source: MarkTechPost
