Mistral Large 4 aka "le Chonk": Europe's Biggest Open Model

Author: Prompt Engineering · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Mistral Large 4 aka “le Chonk”: Europe’s Biggest Open Model · Channel: Prompt Engineering · 01:02 · Watch on YouTube

Topics: AI governance and regulation, AI models, Local and open-source models, Multimodal models

Summary

Mistral has announced Mistral Large 4, nicknamed “le Chonk,” a multimodal mixture-of-experts model with about 1 trillion total parameters and a 1 million token context window. The company says it is competitive with the strongest open models, particularly in coding agents, cybersecurity, finance, and law. The video notes that open-weight models have largely been led by Chinese labs and presents Mistral Large 4 as a notable European entry. It is currently available as a preview API, with weights expected later this month, and its performance remains to be seen.

Key points

Mistral Large 4 is officially nicknamed “le Chonk.”
It is a mixture-of-experts model with about 1 trillion parameters in total and around 50 billion active per token.
The model accepts images as well as text and has a 1 million token context window.
Mistral describes it as frontier-level for an open-weight model.
The company says it is especially strong in coding agents, cybersecurity, finance, and law.
The video names DeepSeek, Qwen, Kimi, and GLM as Chinese labs that have dominated the open-weight space.
Mistral Large 4 is the biggest open model Mistral has made so far.
The model is currently available through a preview API, and Mistral says its weights will arrive later this month.

Chapters

Start	Chapter	Summary
00:00	Mistral announces le Chonk	Mistral Large 4 is introduced, with details on its architecture, scale, modalities, context window, and claimed capabilities.
00:31	A European contender in open weights	The video discusses the model’s place in the open-weight landscape, its preview availability, and the upcoming release of its weights.

Quotes

It’s a mixture-of-experts model with about 1 trillion parameters in total — 00:00

It takes images as well as text and has a 1 million token context window. — 00:00

The open-weight space has been dominated by Chinese labs like DeepSeek, Qwen, Kimi, and GLM. — 00:31

Right now it’s a preview API and Mistral says the weights land later this month. — 00:31

So we’re going to see how actually good this model is. — 01:02

Concepts and entities

Mistral Large 4, Mistral, Mixture of Experts, DeepSeek, Qwen, Kimi, GLM, Reflection, Beam

Full transcript (original language)

00:00 Okay, so Mistral just announced Mistral Large 4, which they are officially calling le Chonk, and the name fits. It’s a mixture-of-experts model with about 1 trillion parameters in total, and around 50 billion of them are active for each token. It takes images as well as text and has a 1 million token context window. Mistral claims it’s frontier-level model in the open-weight category. They say it’s competitive with strongest open models in the world and well ahead of any open model built in the US or Europe, especially on coding agents,

00:31 cybersecurity, finance and law. And that is the most interesting part. The open-weight space has been dominated by Chinese labs like DeepSeek, Qwen, Kimi, and GLM. So it’s great to see a European lab catching up just a day after Reflection released Beam in the US. Mistral has also been fairly quiet at the top end, and this is by far the biggest open model they have ever made. Right now it’s a preview API and Mistral says the weights land later this month.

01:02 So we’re going to see how actually good this model is.