Mistral unveils Large 4: over a trillion parameters, public release on October 27

What happened

According to AIdapted (October 7, 2026), French company Mistral AI presented Mistral Large 4 on October 6, a multimodal open-weight model with 1.05 trillion parameters in total and 49 billion active parameters, in a Mixture-of-Experts architecture. It has a context window of up to 1 million tokens and is designed for tasks such as programming, AI agents, financial analysis, geospatial analysis and chip design. Mistral says it outperforms some Chinese open-weight models in certain areas, including cybersecurity.

Public release is scheduled for October 27, following a testing period.

In plain words

• A 1-million-token context means the model can "keep in view" a huge amount of text at once, like very long documents or conversations.
• Open weights means the model's weights can be made available so it can be deployed and adapted on your own, instead of depending only on a platform.
• Active parameters are the ones the model uses for each response; in a Mixture-of-Experts architecture, only part of the total is activated at a time.

What to do now

The source says public release is scheduled for October 27. If you work with a developer, that is the date to watch.

Sources

• AIdapted: https://aidapted.ro/en/articles/ai-news-october-7-2026-mistral-ai-security/