Mistral Large 4 released on Hugging Face with open weights

Key points

• Multimodal Mixture-of-Experts model, about 1T total and 49B active parameters, with a 1M-token context window.
• Scores 82% on the Artificial Analysis Cyber Index, which tests reproducing and patching a real open-source vulnerability.
• Released under the Mistral Ecosystem Open-Weight License, an MIT derivative that limits model-as-a-service offerings.
• Day-0 support in transformers, vLLM and llama.cpp, plus GGUF quants and TRL fine-tuning, per Hugging Face.

Why it matters for marketers: Teams building security or coding agents can self-host a strong open-weight model instead of depending on closed APIs that may refuse security work.