French AI lab Mistral has released Mistral Large 4 in Research Public Preview, and it immediately tops the Artificial Analysis Intelligence Index for models from outside the United States and China. The 1-trillion-parameter model (49B active) scored 38, comparable to GPT-6 Luna (max, 38) and just behind DeepSeek V4.1 Flash (max, 39).
According to Artificial Analysis, this puts France back at the top of the non-US, non-China AI race, ahead of countries such as South Korea and the United Arab Emirates. Mistral plans to release the model's weights at the end of October.
Mistral Large 4 specs and benchmark results
Beyond raw intelligence, Mistral Large 4 shows strong performance in cyber defense. It scored 50 on the Artificial Analysis Cyber Index, level with GLM-5.3-Flash and ahead of Kimi K3 and DeepSeek V4.1 Flash (max), while its 82% on CyberGym-E2E-AA leads MiMo-V2.6-Pro (79%) and GPT-6 Luna (max, 78%).
The model also improves significantly on document and image reasoning, reaching 19% on GDP.pdf — an 18-point jump from Mistral Large 3. This is partly due to API improvements that now accept 100 images per request, up from 8 in previous Mistral models.
- Intelligence Index: 38 (comparable to GPT-6 Luna max, 38; DeepSeek V4.1 Flash max, 39)
- Cyber Index: 50 (level with GLM-5.3-Flash)
- CyberGym-E2E-AA: 82% (ahead of MiMo-V2.6-Pro and GPT-6 Luna max)
- GDP.pdf: 19% (up 18 points from Mistral Large 3)
