Mistral Large 4 Weights Are Out (Le Chonk): What It Takes to Run

Mistral Large 4 (Le Chonk) open weights went up on Hugging Face on 29 October 2026. We added up every file: 1,060.6 GB for the main FP8 version, 610.0 GB for NVFP4. Mistral's own recipe splits it over 8 GPUs; NVFP4 needs Blackwell; the smallest GGUF is still 233 GB. The API price is unchanged at $0.70 in / $2.10 out per million tokens. Own check (31 Oct 2026): the public Hugging Face API (file list and sizes, config.json, LICENSE) and the public OpenRouter model list, shown as a live terminal recording. We did not run the model and made no paid API call. Benchmark numbers in this video are Mistral's own results, labelled as such. What you see: what was released, why 1 trillion parameters at FP8 is about 1 TB, the GPUs Mistral's and Hugging Face's recipes use, the NVFP4 version, GGUF sizes for llama.cpp, the one license condition (Model-as-a-Service businesses need a commercial license), what changed since the 6 Oct preview, and the API price. Follow-up to our preview video "Mistral Large 4: Open Weights, API Preview, and the Catch":    • Mistral Large 4: Open Weights, API Preview...   Not legal advice: read the license yourself before commercial use. AI narration (Kokoro, voice am_michael). Animations are our own. Sources: https://huggingface.co/mistralai/Mist... https://huggingface.co/mistralai/Mist... https://huggingface.co/api/models/mis... https://huggingface.co/mistralai/Mist... https://huggingface.co/mistralai/Mist... https://mistral.ai/news/mistral-large... https://mistral.ai/news/mistral-large-4/ https://docs.mistral.ai/models/mistra... https://huggingface.co/api/models/mis... https://huggingface.co/blog/mistral-l... https://huggingface.co/ggml-org/Mistr... https://huggingface.co/unsloth/Mistra... own check (Hugging Face API + OpenRouter public model list, no key) Chapters: 00:00 The weights are a download now 01:35 Our own check: every file, added up 03:30 What hardware Mistral's own recipes use 05:16 Can you run it at home? 07:28 What changed since the preview 09:26 Should you download it? Visual credits: Hugging Face repository page capture (mistralai/Mistral-Large-4-1T-A52B) and official Mistral mark in original animation https://huggingface.co/mistralai/Mist... Hugging Face repository page and Mistral Large 4 model card captures under commentary https://huggingface.co/mistralai/Mist... Own live terminal recording (Hugging Face API + OpenRouter public model list, 31 Oct 2026) https://huggingface.co/mistralai/Mist... Original Claude-Design animation of the mixture-of-experts memory point Original Remotion animation of model-card and Hugging Face blog recipes plus our arithmetic Original Remotion animation of NVFP4 model-card facts plus our arithmetic Original Remotion animation of GGUF model-card sizes Mistral Ecosystem Open-Weight License passages (LICENSE file on Hugging Face) under commentary https://huggingface.co/mistralai/Mist... Mistral release post (29 Oct 2026) passage captures under commentary https://mistral.ai/news/mistral-large... Mistral release post (29 Oct 2026) price passage capture under commentary https://mistral.ai/news/mistral-large... Mistral Large 4 model card passage capture in original animation https://huggingface.co/mistralai/Mist... Original HyperFrames summary animation I'm Ronald, founder of Datawonder. I build AI systems, voice agents, and custom software for small and midsize businesses, and on this channel I look at what AI can actually do for your company, past the demos and the noise. 🔗 Connect Book a free 30-minute call: https://calendar.app.google/KvLN18zmK... Website: https://datawonder.nl/en/ LinkedIn: / ronald-postelmans Instagram: / ronaldairocks X / Twitter: @Ronald999283845 TikTok: / ronalddatawonder AI-generated narration. Original analysis and illustrative graphics; source demonstrations are attributed.