On October 29, 2026, Mistral released the open weights of Mistral Large 4, a model the company nicknames "Le Chonk". That means anyone can download the full model from Hugging Face and run it on their own hardware, not just rent it through an app. It's one of the biggest open-weight releases of the year, and it raises a fair question for everyday AI users: what does "open weights" actually change?
This guide explains Mistral Large 4 in simple terms: what it is, what the licence allows, what hardware it needs, and who should care. We've used only Mistral's and Hugging Face's own pages for the facts.
QUICK ANSWER
On October 29, 2026, Mistral released the open weights of Mistral Large 4, nicknamed Le Chonk. Anyone can download the full model from Hugging Face and run it on their own hardware. It is a natively multimodal Mixture-of-Experts model with about 1 trillion total parameters, so it needs serious hardware, not a laptop.
What is Mistral Large 4?
According to the official model card, Mistral Large 4 is Mistral's "largest and most capable model to date": a model with about 1 trillion total parameters, natively multimodal, built as a Mixture-of-Experts. It was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own European datacenters.
Key facts from the model card:
- Size: ~1T total parameters, with 49B active per token (52B including embeddings and output layers).
- Context: a 1M-token context window.
- Input: text and images in, text out.
- Languages: natively fluent in 160+ languages, including every official EU language.
- Reasoning modes: low, medium and high (high is the default).
Mistral first released Large 4 as a public preview on October 6, and the open weights are an improved version of that preview (Mistral announcement).
Open weights explained simply
Most popular assistants, like ChatGPT, Gemini and Claude, are "closed". You use them through the company's app or API, and the model files stay on the company's servers. With an open-weight model, the trained model files are published. You can download them, run them privately, fine-tune them on your own data, and build products without sending your data to anyone else.
Open weights are not the same as fully open source. You get the trained model, but not necessarily the training data or the whole recipe. And the licence still sets rules, which brings us to the next point.
What does the Mistral Large 4 licence allow?
Mistral releases the model under the Mistral Ecosystem Open-Weight License, which Mistral describes as a modified MIT licence. In Mistral's words, the model is "free to use, modify, fine-tune, and commercialize, with a single condition": businesses that operate a Model-as-a-Service offering must get a separate commercial licence from Mistral first.
So a startup can build a Mistral-powered feature into its own app. A company that wants to resell access to the model itself as a hosted AI service needs to talk to Mistral. If you plan anything commercial, read the licence on the model card yourself rather than relying on a summary.
Can you run Mistral Large 4 on your laptop?
Realistically, no. This is a trillion-parameter model. Hugging Face's own deployment recipe uses 8 NVIDIA H200 GPUs (Hugging Face blog), and SGLang's verified recipes use 8× B300 for the FP8 weights or 4× GB300 for the smaller NVFP4 version (SGLang docs). GGUF builds exist from Unsloth and the llama.cpp team, but even compressed, this is server-class hardware.
For most readers, the practical ways to try it are:
- Mistral Studio or the Mistral API (model name
mistral-large-4). - OpenRouter, which Mistral lists as a launch option.
- Hugging Face Inference Endpoints or Inference Providers, if you want a hosted copy.
Mistral Large 4 price on the API
Mistral lists API pricing at $0.70 per million input tokens, $0.07 per million cached input tokens, and $2.10 per million output tokens. For comparison, OpenAI lists GPT-6.1 Terra at $1 input and $5 output per million tokens, which we covered in our GPT-6.1 Terra explainer. Price alone doesn't decide quality, but it's a useful reference for small teams.
How good is Mistral Large 4?
Here's what the companies themselves report, so read these as claims:
- Hugging Face says that as of October 29, 2026 it was the fifth-best open-weights model on the Artificial Analysis Intelligence Index.
- On coding, Hugging Face lists 61.7% on DeepSWE v1.1 and 28.3% on Terminal-Bench 4.
- Mistral's model card says it shows especially good performance on cybersecurity, finance and law, and "surpasses frontier closed models on visual grounding".
If you are new to the GPT-6 family, our GPT-6 explainer covers the closed-model side. In short: strong for an open model, not a claim to beat the best closed models across the board. If you mainly want a chat assistant, the closed options in our ChatGPT vs Gemini vs Claude guide are still easier to use.
Why the cybersecurity note matters
Mistral pitches Large 4 as "Built for Coders, Trusted by Defenders" and says it is state of the art among open-weight models for security work while refusing malicious requests. Open models are a double-edged sword here, because once weights are public, anyone can run them. That's why it's worth watching how Mistral updates guardrails over time, and why businesses running it should apply their own safety checks.
Who should care about Mistral Large 4?
Developers and startups
If you need data to stay on your own servers, or want to fine-tune on private documents, an open-weight model at this level is a big deal.
Businesses in Europe and beyond
Mistral frames this as "sovereign AI": a European-trained model you can host yourself. That appeals to organisations with strict data rules.
Students and everyday users
You probably won't download it. But open releases push prices down and give app makers more choice, which tends to reach you through cheaper, better tools.
Want to put this into practice? Try our free AI API Cost Calculator and AI Token Counter right in your browser, then read Nvidia Hugging Face Deal and ChatGPT vs Gemini vs Claude (2026) on Botvana.
Related: Curious about a different way to build language models? Read what a diffusion language model is (Clover-35B explained).
Sources and further reading
- Mistral-Large-4-1T-A52B model card (Hugging Face)
- Mistral Large 4 (Le Chonk) is now open weight (Mistral AI)
- Welcome Le Chonk: Mistral Large 4 on Hugging Face (Hugging Face)
- Mistral Large 4 (SGLang)
FAQ
When were Mistral Large 4's weights released?
October 29, 2026, after a public preview on October 6.
Is Mistral Large 4 free?
The weights are free to download and use under Mistral's licence. Running them needs expensive hardware, and the API is paid.
What does "Le Chonk" mean?
It's Mistral's nickname for Mistral Large 4.
Can I use Mistral Large 4 commercially?
Yes, with one condition: Model-as-a-Service businesses need a separate commercial licence from Mistral.
What is Mistral Large 4's context window?
1 million tokens.
Comments
Post a Comment