Mistral unveils Le Chonk, a large open-weight model you can run on your own servers

Mistral has announced Le Chonk, officially Mistral Large 4, an open-weight model for private deployment, due on October 27 after a €3 billion raise.

In 30 seconds French company Mistral has announced Le Chonk, officially Mistral Large 4, its biggest open-weight model. It arrives on October 27, and you can download it and run it on your own hardware. Mistral says it beats many open-weight rivals, but independent testing has yet to confirm that.

What happened

On Tuesday, Mistral announced the coming release of Le Chonk, as reported by DW. It is the company's largest open-weight model and is aimed at users who want to keep their data private. It will be publicly available on October 27, and it is Mistral's first new model since May.

Open weights mean you can download the model's parameters, customise them and run the software on your own servers. That is different from the closed models offered by Anthropic and OpenAI, where the weights stay with the provider. Many Chinese competitors, by contrast, also publish open weights. Before the public launch, cybersecurity experts and government authorities will get a version with fewer safety restrictions so they can test what it can do.

Mistral is widely regarded as Europe's leading AI company. Image: Unsplash — three black spray bottles on a black surface

The narrative that Europe cannot compete is something that is not true.

Arthur Mensch, Mistral CEO, speaking at the company's event in Abu Dhabi

Why it matters

Self-hosting is the main selling point. If the weights run on your own infrastructure, prompts and documents never leave your network, which matters for regulated sectors and for anyone wary of sending data to a US provider. Mistral has positioned itself as a safer European alternative, and last month it raised €3 billion, described as Europe's biggest technology fundraising.

Mensch said the model is "actually above the Chinese models" in certain areas, including cyber. He did not say which models he meant, and the report gives no benchmark figures. We would treat that as a claim to verify, not a result. The limited-restriction preview for security experts and authorities is a useful sign that outside testing is planned.

What we don’t know yet

• The hardware needed to run Le Chonk on your own servers.
• The licence terms attached to the downloadable weights.
• Which Chinese models Mistral compared itself against, and on what benchmarks.
• How much the reduced-restriction version differs from the public release.

Our recommendation is simple: if private deployment matters to you, plan a test for after October 27 and compare it against your current model on your own tasks. Until then, the announcement tells us what Mistral intends, not what the model delivers.

Further reading: supports HTML5 video · Preferred Source on Google by clicking here.

Opinion · If I were…

If I were a company weighing Le Chonk, I would wait for the October 27 release and test it myself. The CEO's claim that it beats Chinese models in cyber came without named rivals or numbers. A claim without a benchmark is marketing.

I would also check what hardware it needs before getting excited about private deployment. Running it yourself means paying for the servers yourself.

Related articles from LaRebelión

South Korea bets $3.5 billion on its own frontier AI modelGoogle pauses its open-source bug bounty as AI slop floods maintainersAI agents create fresh cybersecurity risks and liability dilemmas

Original source: DW.com

Produced with AI support and reviewed by the newsroom

Falcon · Signals analyst · Riyadh

“Open weights sell privacy, but a vague claim of beating Chinese models proves nothing until someone runs the tests.”