Mistral AI Updates: October 30, 2026

1. Mistral Large 4 Open Weights Land on Hugging Face

Mistral AI released the weights for Mistral Large 4 (“Le Chonk”), its multimodal Mixture-of-Experts model that has been in API public preview since October 6. The weights come in FP8 (mistralai/Mistral-Large-4-1T-A52B) and NVFP4 checkpoints under the Mistral Ecosystem Open-Weight License, an MIT derivative that restricts model-as-a-service offerings. Large 3 used Apache 2.0. The model has about 1T total parameters and 49B active. It uses 72 layers with Multi-head Latent Attention and a sparse attention indexer, routes each token to 4 of 128 experts plus a shared expert, includes a 1.6B Pixtral vision encoder, and supports a 1M-token context. Reported scores include 61.7% on DeepSWE v1.1, 28.3% on Terminal-Bench 4 and 93% on 40 Cybench challenges. Artificial Analysis ranks it the fifth-best open-weights model on its Intelligence Index. Day-0 support covers Transformers, vLLM (NVFP4 requires Blackwell GPUs), llama.cpp with GGUF quants, TRL LoRA fine-tuning and Hugging Face Inference Endpoints. Reasoning effort is set to low, medium or high and cannot be turned off. Source