2026-10-07 / Signal #5
Mistral dropped Large 4 (“Le Chonk”), a 1-trillion-parameter European open-weight model
“France’s new 1-trillion-parameter model is nicknamed Le Chonk and the weights are coming.”
The Story
Mistral launched a public preview of Large 4 (unofficially “Le Chonk”), a 1T-parameter natively multimodal MoE with 49B active parameters, trained from scratch on ~3,800–4,000 NVIDIA Grace Blackwell GPUs in its own European data centers. Weights promised by end of October. Strong coding, cyber, and some visual-grounding claims; positioned as a European “third way” with 160+ languages including all official EU languages.
Why It Matters
Europe just tried to drop a chonky open frontier model trained on its own iron. Builder utility (weights incoming) plus the slightly absurd national-industrial vibe of naming a trillion-parameter model after a chunky animal. Not generic US-lab PR.
Evidence
Mistral official post, VentureBeat, Artificial Analysis.
Sources
Daily scan: 2026-10-07