Welcome Le Chonk: Mistral Large 4 on Hugging Face
- ID
- 40768
- Status
- summarized
- Published
- 29 Oct 2026, 8:00 AM
- Fetched
- 29 Oct 2026, 11:23 PM
- Provider
- Hugging Face Blog
- Category
- developer-ai
- Original URL
- https://huggingface.co/blog/mistral-large4
- Source URL
- https://huggingface.co/blog/feed.xml
Summary
- Score
- 6.5
- Created
- 29 Oct 2026, 11:24 PM
- Tags
- Audience
- developersai_ml_learnersai_agent_userssaas_startup_founders
What happened
Mistral released Large 4 ('Le Chonk'), a multimodal Mixture-of-Experts model with about 1T total and 49B active parameters, a 1M-token context window, and text+image input, under the Mistral Ecosystem Open-Weight License (an MIT derivative with limits on model-as-a-service). It ships as FP8 and NVFP4 checkpoints plus Unsloth/llama.cpp GGUFs, with support in Transformers, vLLM, llama.cpp, TRL fine-tuning, Inference Endpoints, and Dell Enterprise Hub. Mistral reports 82% on the Artificial Analysis Cyber Index—reproducing and patching a real open-source vulnerability—and says several closed frontier models score near zero because they refuse the task.
Why it matters
If you build coding agents or defensive-security tooling that needs to inspect vulnerable code and patch it, Le Chonk is a concrete open-weight option to test locally via vLLM or llama.cpp instead of depending on an API that may refuse the work or cut access mid-incident. Check the model-as-a-service license before hosting it for customers, and treat the 82% cyber score as vendor-reported until you reproduce it on your own vulnerability-repro task.
Discussion angle
Is an open-weight model that will reproduce vulnerabilities more useful for defenders than a closed API that refuses—and does the model-as-a-service restriction kill the obvious SaaS wrapper?