The weight file is still a countdown
On October 6, 2026 Mistral put Large 4, also called Le Chonk, into a public preview on Mistral Studio. That post calls it a 1 trillion-parameter model with 52 billion active parameters, trained on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own datacenters in Europe. It says weights drop at the end of this month, and later that the weights will be released by the end of the month. The post does not link a weight file.
On October 21, 2026 a Mistral post about work with NVIDIA describes that same preview as approximately 1 trillion total parameters and 49 billion active parameters, trained on 3,800 NVIDIA Grace Blackwell GPUs in Mistral-operated data centers in Europe, with weights planned for release at the end of October. The mistralai Hugging Face card for the upcoming release says the model has 49 billion active parameters per token and 52 billion including embeddings and output layers. It lists October 31, 2026 as the current ETA, says it is waiting for the release, and names a planned artifact, mistralai/Mistral-Large-4-1T-A52B. No weight file was downloaded for this update.
A China test quote, not a new checkpoint
China Daily reported on October 15, 2026 that Apple executive chairman Tim Cook said Apple Intelligence has been approved for the Chinese market and that a test version will be released later this month. He named real-time translation, writing tools, and image cleanup as the first features, and said the AI-powered Siri will not be in that initial rollout. The article says license records updated on September 29 added three filings, including Alibaba's Qwen, numbered Shanghai-AppleZhiNeng-202506160057, Shanghai-TongYiWanXiang-202410090024, and ZheJiang-TongYiQianWen-20230901. No Apple or Alibaba page opened for this update shows that test as available, and the article does not name a Qwen checkpoint.
Z.ai's public list still ends in August
Z.ai's new-released notes still put the newest model entries on August 26 and August 18, 2026. The August 26 entry describes a hybrid model with 320 billion total parameters and 18 billion activated. The August 18 entry names GLM-5.3 and says it gained 50 percent over GLM-5.2 on Z.ai Code Bench. The page does not name GLM-6.0 and has no model date after August 26, so it does not add a model a customer can call.
What to watch through October 31
Watch whether a Large 4 weight file is public by the October 31 date on the Hugging Face card, and whether its active-parameter count matches that card, the October 6 post, or the October 21 post. Watch whether the China test Cook described starts, and which Qwen checkpoint it uses. Watch whether Z.ai adds a model entry newer than August 26.
Humanoid shipments, robotaxi city lists, electric-aircraft orders, and product news from companies outside the frontier-model race do not change which model a customer can call while Large 4's weights are still listed as waiting and Z.ai's notes still end in August.
























