Hugging Face model card for mistralai/Mistral-Large-4-1T-A52B showing the model description and quantizations
Mistral Large 4 model card on Hugging Face, captured 2026-10-30. Image: AppFor.ai screenshot of Hugging Face's page. Image source ↗

What happened

Mistral published the weights of Mistral Large 4, nicknamed Le Chonk, on October 29, 2026, three weeks after the paid public preview of October 6. Mistral says the released version had three more weeks of reinforcement learning, with gains in coding and defensive security and fewer output tokens per task.

Reasoning now has three settings: low, which replaces the old none mode, medium for everyday work and high, the default, for the hardest tasks. The model card recommends temperature 1.0 and top_p 0.95 at every setting.

How to download and run it

The official checkpoint is mistralai/Mistral-Large-4-1T-A52B on Hugging Face in FP8, plus an NVFP4 version for Blackwell GPUs with a smaller memory footprint. Hugging Face's launch post adds GGUF quantizations from Unsloth and the llama.cpp team. The card counts 49 billion active parameters per token, or 52 billion including embeddings and output layers.

Mistral recommends vLLM: pull the vllm/vllm-openai:mistral-large-4 image and serve with --tensor-parallel-size 8, --tool-call-parser mistral and --reasoning-parser mistral, the last two needed for tools and separate reasoning output. SGLang has its own lmsysorg/sglang:mistral-large-4 image. A built-in DSpark draft head enables speculative decoding, which Mistral measured at up to 1.9 times the throughput.

What it means for you

The licence is the Mistral Ecosystem Open-Weight License, a modified MIT licence that allows commercial use and fine-tuning. The one condition: a company that operates a Model-as-a-Service business needs a separate commercial licence from Mistral before using the model commercially.

Without the hardware, start on the Mistral API, where the model is mistral-large-4, or on the demo Spaces Mistral funds with a small free budget during the first week. Self-host once your tests show it handles your tasks and data residency requires it.

Sources & further reading

Based on the official documentation below. Features and access can change; check the provider before making a decision.

  1. Mistral AI: Mistral Large 4 is now open weight (opens in a new tab)
  2. Hugging Face: Mistral-Large-4-1T-A52B model card (opens in a new tab)
  3. Hugging Face blog: Welcome Le Chonk (opens in a new tab)
  4. Mistral Ecosystem Open-Weight License (opens in a new tab)
Something out of date or incorrect?Report a correction

Was this page useful?

Help us make our coverage more useful. This is feedback on the page, not a rating of the tool.

Share this page

Related reading