What actually happened

Mistral launched a public preview of Mistral Large 4 on 6 October 2026, nicknamed Le Chonk. On the company's own page it is a 1 trillion-parameter natively multimodal model with 52 billion active parameters, the largest model Mistral has trained. Some secondary write-ups put active parameters at 49 billion. Use the primary figure until Mistral publishes the architecture note.

The preview API is available in Mistral Studio now. Weights are not. Mistral says they drop by the end of the month, after red-teaming with cybersecurity leaders, vetted partners, and state authorities. The model was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own European datacenters, and the preview is served on that same infrastructure. Mistral positions it as state of the art among open models on cybersecurity, finance, and law, and ahead of closed models on some visual-grounding tasks. Those are the company's results. Broader independent indexes published in the first days sit nearer everyday frontier models than the top closed systems, so the leaderboard story is still incomplete.

Why a small team cares

A studio that has been asked for an EU-trained, eventually self-hostable model finally has a frontier-scale name to put in the proposal. That is the positive shift. The Monday constraint is the same one Mistral states: weights later this month, preview API today. You can show a client a live European endpoint this week. You cannot hand them a weight file, a license, or a private-cloud run yet.

Hype vs useful

The useful headline is a European frontier candidate with a date on the weights, not a model you cut over to today. "Best open model outside China" and the cyber scores are Mistral's framing, and several closed models refuse the cyber tasks those charts reward. Wait for the weight drop, the license, and one independent score on a workload you care about. Until then the preview is a demo you can run, and a reason to keep the sovereignty conversation open.

Source: mistral.ai (https://mistral.ai/news/mistral-large-4/)