Le Chonk arrives as Mistral launches Large 4
mistral
| Source: Mastodon | Original article
French AI firm Mistral has unveiled its new flagship model Large 4, dubbed Le Chonk, with a trillion parameters.
Mistral AI has moved its flagship model out of preview and into public hands. On 6 October the French company released a preview of Mistral Large 4 – dubbed “Le Chonk” – a mixture‑of‑experts (MoE) system that tops one trillion parameters in total, with roughly 49 billion active weights at inference time. The preview is already accessible via API, and the full weight files are slated for open release on 27 October after a final safety audit.
The launch matters because it marks the first open‑weight, trillion‑parameter model to emerge from Europe. Earlier this week we noted Mistral’s preview of the same model, but today the company confirms that the weights will be publicly available, a step that could level the playing field with larger, closed offerings from the United States and China. Le Chonk is also multilingual by design, having been trained on more than 160 languages, including all 27 official EU languages, which positions it as a versatile tool for developers building applications across the continent.
What to watch next is the impact of the open‑weight release. Developers will be able to fine‑tune the model locally, and the announced API pricing will shape adoption rates. Benchmarks such as DeepSWE and Cybench, already hinted at in the preview, will be scrutinised for real‑world performance. Regulators and AI safety groups are likely to focus on the upcoming safety review, while competitors may accelerate their own open‑weight roadmaps. The October 27 deadline therefore becomes a key date for Europe’s AI ecosystem, signalling whether open, large‑scale models can gain traction against proprietary alternatives.
Sources
Back to AIPULSEN