Mistral Releases Large 4 — A Trillion-Parameter Multimodal MoE Model

On October 6, 2026, Mistral AI opened a public preview of the multimodal Mistral Large 4 model — a Mixture-of-Experts with 1 trillion parameters and 49 billion active, a 1 million token context, trained on 3,800 NVIDIA Grace Blackwell GPUs in its own European data centers, with access via API in Mistral Studio and a promise of open weights by the end of October.

Compiled by Sergey KostenchukPublished 2026-10-06Updated 2026-10-06

2026-10-06 Research Anthropic

🤖 Mistral Large 4, also known as “Le Chonk”: a trillion-parameter model

Mistral AI opened a public preview of the multimodal Large 4 model: a MoE with 1 trillion parameters (49 billion active), a 1 million token context, and a 1.6 billion vision encoder. Trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its own European data centers; open weights are promised by the end of October.

🌍 Mistral claims leadership among open weights outside China: 82% on CyberGym-E2E (Claude Opus 5.5 and GPT-6 Astra refuse to perform this task), 93% on Cybench. The figures are currently vendor-run, and the license terms have not been announced.

👤 The API is already available in Mistral Studio at $1.36/$4.18 per 1 million input/output tokens — you can compare it with Claude and GPT on your own tasks, and after the weights are released — run the model locally.

Source 1: https://mistral.ai/news/mistral-large-4/ Source 2: https://techcrunch.com/2026/10/06/mistrals-new-1t-model-aims-to-leapfrog-closed-and-open-rivals/

Sources