Mistral Large 4 lands on Hugging Face: 1T-param MoE with 1M context
mervenoyann · x · 2026-10-30
Mistral released Mistral Large 4 (aka Le Chonk), a multimodal MoE model with 1T total / 49B active parameters and a 1M-token context window. Key points:
- Ships in FP8 and NVFP4, plus community GGUFs from Unsloth and the llama.cpp team
- Mistral Ecosystem Open-Weight License (MIT derivative restricting model-as-a-service)
- Sparse attention plus a Pixtral vision encoder; notably strong at cybersecurity, coding agents, and multimodality
- Day-0 support in transformers, vLLM, and llama.cpp, with TRL fine-tuning available
More from Models
- Altman: a new OpenAI model 'knocked down' research problems, Navier-Stokes proof cost ~$10M — chrismattmann · 2026-10-30
- xAI has everything it needs, so why is it falling behind at the frontier? — Snoo26837 · 2026-10-30
- Researchers Observe Opus 3's Simulated Characters Almost Always Share Its Own Values — repligate · 2026-10-30
- evident-charts: open-source skill teaches coding agents to make honest charts and self-check — randal_olson · 2026-10-30
- GPT 6 Sol vs Claude Opus 5.5 low: same prompt, black hole sim showdown — opmgyhx · 2026-10-30
- Elorian launches visual reasoning model Glimpse alongside DecoyCount benchmark — AndrewDai · 2026-10-30