Mistral Large 4 arrives with open weights and a 1M-token context
Mistral’s multimodal MoE model is now on Hugging Face in FP8 and NVFP4, combining text and image inputs with a one-million-token context.
AT A GLANCE
Key points
- The model has about 1T total and 49B active parameters.
- It accepts text and images with a one-million-token context.
- FP8, NVFP4 and community GGUF options are available.
- Transformers, vLLM and llama.cpp have day-zero support.
What is changing?
Hugging Face released the open weights for Mistral Large 4 on October 29. The multimodal, multilingual Mixture-of-Experts model, codenamed “Le Chonk,” takes text and images as input and generates text. The announcement lists about one trillion total parameters, 49 billion active parameters and a one-million-token context window.
Official FP8 and NVFP4 variants are available on Hugging Face, alongside GGUF builds prepared by llama.cpp and Unsloth. Day-zero support covers Transformers, vLLM and llama.cpp, while the Inference Endpoints recipe runs on eight H200 GPUs and serves contexts up to 131K tokens.
Mistral positions Large 4 for coding, cybersecurity and multimodal work. The weights use the Mistral Ecosystem Open-Weight License, which places limits on model-as-a-service offerings; open weights therefore does not mean unrestricted commercial use.
Explore the original sources
Source date: 29 October 2026.
Published by AI Radar: 29 Oct 2026, 18:07 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…