88Z AI, 8 October — Mistral has released its flagship model, Mistral Large 4, featuring 1 trillion parameters and a mixture-of-experts architecture. In third-party evaluations, its intelligence index matches GPT-6 Luna, outperforming GPT-6 and Claude in several rigorous tests, with an 82% cybersecurity vulnerability fix rate. Trained on 3,800 Blackwell chips and supporting over 160 languages, the model's weights will be fully open-sourced by the end of the month.
At a glance
- Mistral Large 4 has 1 trillion parameters, matches GPT-6 Luna in intelligence, and will have its weights fully open-sourced by month-end.
- Trained on 3,800 Blackwell chips, the model supports over 160 languages and achieves an 82% cybersecurity vulnerability fix rate.
The update
Mistral has officially launched its latest flagship large model, Mistral Large 4. The model features a total of 1 trillion parameters and utilizes an advanced mixture-of-experts architecture. According to third-party evaluations, Mistral Large 4's intelligence index is on par with GPT-6 Luna, and it outperforms both GPT-6 and Claude in several rigorous tests. Furthermore, it achieves an 82% success rate in fixing cybersecurity vulnerabilities. Regarding training infrastructure, the model was trained using 3,800 Blackwell chips, providing a robust computational foundation. Mistral Large 4 supports over 160 languages worldwide, demonstrating broad applicability. Notably, Mistral plans to fully open-source the model's weights by the end of this month, making them freely available for developers and researchers.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.