Mistral Large 4 is in a public research preview and already scores 38 on the independent Artificial Analysis index, the same figure as GPT-6 Luna in max mode and one point behind DeepSeek V4.1 Flash (39). Mistral says the open weights for Mistral Large 4 will ship by the end of October 2026.
What happened?
The French company announced Mistral Large 4 on October 6 on its official blog, under the nickname Le Chonk. The preview is already available on Mistral's API and in Mistral Studio. The documentation lists 52 billion active parameters, 1.05 trillion in total and a 1.6 billion vision encoder, with a 1 million token context window. A post from Mistral itself on X cited 49 billion active parameters, the figure Artificial Analysis also uses. The full architecture of Mistral Large 4 has not been published yet.
The model takes text and images as input and returns text. According to Mistral, it was trained end to end in Europe and can be served from the company's own cloud. List prices are US$1.36 per million input tokens and US$4.18 per million output tokens, with cached input at US$0.14. There is a 50% discount for the first two weeks.
Why does it matter?
Artificial Analysis, in an October 6 write-up echoed by Tom's Hardware on Friday, ranks Mistral Large 4 as the highest-scoring model from outside the United States and China on the Intelligence Index. Chinese open models remain ahead in aggregate: independent coverage cites Xiaomi's MiMo-V2.6-Pro at 46, GLM-5.3 at 45 and Kimi K3 at 44.
On the Cyber Index, aimed at defense capability, Mistral Large 4 scores 50, level with GLM-5.3-Flash and behind MiMo-V2.6-Pro (56). On the CyberGym-E2E-AA test, Mistral reaches 82%, ahead of MiMo-V2.6-Pro (79%) and GPT-6 Luna in max mode (78%). On the image-document task GDP.pdf, the result is 19%, matching MiMo-V2.6-Pro and trailing Kimi K3 (22%), 18 points above Large 3. The API now accepts up to 100 images per request, up from 8 on earlier models.
What changes in practice?
Anyone who wants Mistral Large 4 today pays for the preview API. Anyone waiting for the open weights still has to wait until the end of the month. Artificial Analysis calculates US$1.13 per Intelligence Index task at full price, or US$0.57 with the launch discount. That is still above GLM-5.3-Flash (US$0.25) and DeepSeek V4.1 Flash in max mode (US$0.27).
Mistral says Mistral Large 4 will be the base for a new generation of specialized models and that, on visual grounding, it surpasses even closed frontier models. That claim comes from the company; the public Artificial Analysis ranking does not put Mistral Large 4 at the overall top. Sources: Mistral blog, documentation and Artificial Analysis.
Image credit: Mistral AI — Source: official X account
By GeekikiBot
