What Is Mistral Large 4? “Le Chonk,” the 1T Open Model, Explained

Mistral Large 4 is the biggest AI model the French company Mistral has ever built, and it has a nickname that sounds like a joke: “Le Chonk.” It is back in the spotlight this week as developers count down to the day its weights become free to download.

So what is it, can you use it today, what does it cost, and is it really as good as the American and Chinese models? Here is the short, plain-English version.

What is Mistral Large 4?

Mistral Large 4 (also written Mistral Large 4.0, ML4 or “mistral-large-4”) is Mistral AI’s new flagship model. Mistral calls it a natively multimodal, mixture-of-experts model. That last part means the model is split into many small “experts,” and only a few of them wake up for each word. So it has about 1 trillion parameters in total, but only about 49 billion are used per token, or 52 billion if you count the embedding and output layers, per its Hugging Face card.

Mistral says it trained the model from scratch on 3,800 Nvidia Grace Blackwell GPUs in its own European data centers. It is the first model from the €3 billion funding round Mistral closed in September, according to The Next Web.

Why is it called Le Chonk?

It started as a meme. Earlier this year, fans on Reddit and X joked about an imaginary giant Mistral model, and CEO Arthur Mensch played along. Mistral’s launch post now lists the model, “very officially,” as le Chonk. “Chonk” is internet slang for a big, round cat.

Mistral Large 4 at a glance

Mistral Large 4 release date: when do the weights drop?

The model itself is already out as a preview. What people are waiting for is the open-weight release, which lets anyone download the model and run it on their own servers. Mistral’s post says this is planned for the end of October 2026. The Hugging Face page lists October 31 as the current ETA, while The Next Web reported October 27. Until Mistral posts the files, treat any exact day as a target, not a promise.

Before the weights go out, Mistral says it is red-teaming the model with cybersecurity firms, vetted partners and government agencies. Those groups get a version with lighter moderation and stronger cyber skills than the public API. Mistral also promised more details on the architecture, benchmarks and safety testing.

Mistral Large 4 benchmarks: how good is it?

Every number below comes from Mistral and is preliminary, so read it as a sales pitch until independent tests catch up.

• Coding: 61.7% on DeepSWE v1.1 and 28.3% on Terminal-Bench 4. In a human coding review by Surge AI, it ranked second of five models.
• Cybersecurity: 93% on Cybench and 82% on a test where the model must reproduce and patch real vulnerabilities.
• Images: 42% on Dense 200, a hair ahead of GPT-6 Astra’s 41%.
• Safety: it resisted 93.3% of attacks on Lakera’s B3 benchmark.

Mistral calls it the strongest open-weight model built outside China. Independent testing paints a more mixed picture: Decrypt reports that early Artificial Analysis results put it well behind the top closed models from OpenAI and Anthropic, while still ahead of the best US open-weight model. Vals AI ranks it #5 of 24 on one of its indexes.

Mistral Large 4 vs Large 3 and Chinese open models

On Mistral’s own charts, Large 4 edges out DeepSeek-V4-Pro and GLM-5.3 on DeepSWE coding and ties DeepSeek-V4-Pro on a finance test, per The Next Web. DeepSeek also stayed busy this month with DeepSeek V4.1 Flash, a much smaller and cheaper model.

Is Mistral Large 4 free?

Not today. The preview is a paid API, though you can test it in the Mistral Studio playground. Once the weights are out, downloading them should cost nothing, but running a 1-trillion-parameter model needs a rack of data center GPUs, not a laptop. For most people, “free” will mean a hosted version through Mistral or a provider like OpenRouter. Mistral has not said when Large 4 comes to its Le Chat app.

What this means for you

Le Chonk matters most to companies and governments that want a top model they can run on their own machines, in Europe, without anyone being able to cut off access. That is Mistral’s whole pitch, and it explains why cyber defense is front and center.

For a normal person on a phone, one more model name is mostly noise. The useful question is not “which model wins this week?” but “can something just do the task for me?” That is the idea behind agentic AI: you describe the job, and agents plan it and finish it. Agentic AI gives you a team of up to four expert agents on iPhone and iPad, with no account needed, so you can skip the model shopping. Mistral’s other launches this month, Voxtral 3 for voice and Timestral for forecasting, show the same trend: specialists working together.

Quick answers

What is Mistral Large 4? Mistral AI’s flagship open-weight model, nicknamed Le Chonk, with about 1 trillion parameters and support for text and images. It entered public preview on October 6, 2026.

When is the Mistral Large 4 release date for open weights? Mistral says the end of October 2026. Its Hugging Face page shows October 31 as the ETA; the exact day is not confirmed.

How much does Mistral Large 4 cost? $1.36 per million input tokens, $0.14 per million cached input tokens and $4.18 per million output tokens on the Mistral API.

Is Mistral Large 4 on Hugging Face? Its page exists but is marked “upcoming release.” The files are not downloadable yet.

Is Mistral Large 4 open source? Mistral calls it open-weight, but it has not named the license yet. Mistral Large 3 used Apache 2.0.

Mistral Large 4 vs GPT-6 Astra? Mistral claims wins on some image, finance and legal tests. Early independent tests suggest top closed models from OpenAI and Anthropic are still ahead overall.

New to words like “parameters” or “open weights”? Our agentic AI glossary explains them in plain English.

Sources: Mistral: Mistral Large 4 · Mistral docs · Hugging Face model card · The Next Web · Decrypt · Vals AI