Mistral Large 4 arrives in research preview at 38 on the Intelligence Index, matching US and Chinese frontier models
Synopsis
Mistral released Mistral Large 4 as a Research Public Preview, a 1T-parameter model with 49B active parameters whose weights are planned for release at the end of October. It scores 38 on the Artificial Analysis Intelligence Index, comparable to GPT-6 Luna (max, 38) and just below DeepSeek V4.1 Flash (max, 39), making it the highest-scoring model from outside the US and China. The same preview scores 50 on the Cyber Index, level with GLM-5.3-Flash, while costing $1.13 per Intelligence Index task at standard pricing, above similar-intelligence open-weights models.
Interpretation
Mistral released a Research Public Preview of Mistral Large 4, returning France to the position of holding the most intelligent model from outside the US and China. The comparison is against other frontier models on the same index: DeepSeek V4.1 Flash (max) scores 39 and GPT-6 Luna (max) scores 38, while the Mistral Large 4 preview scores 38, level with the latter. The ranking rests on a score of 38 on the Artificial Analysis Intelligence Index, with South Korea and the United Arab Emirates named among the countries it moves ahead of.
The model takes text and images as input and returns text, with a 512k-token context window, available through Mistral's API. Weights are not yet open; the plan is to release weights for the 1T-parameter, 49B-active model at the end of October. These details come from the context window, multimodality and availability listed on the model page and release note.
On cyber defense, the preview scores 50 on the Artificial Analysis Cyber Index, level with GLM-5.3-Flash and behind MiMo-V2.6-Pro at 56. Once its weights are released, it will rank among the top three open-weights models on that Cyber Index. Its strongest single result is CyberGym-E2E-AA at 82%, ahead of MiMo-V2.6-Pro at 79% and GPT-6 Luna (max) at 78%.
Standard pricing is $1.36/$4.18 per 1M input/output tokens and $0.14 per 1M cached input tokens, giving a cost of $1.13 per Intelligence Index task. A 50% launch discount for the first two weeks brings the cost per task to $0.57, still above GLM-5.3-Flash at $0.25 and DeepSeek V4.1 Flash (max) at $0.27. On document and image reasoning, the preview scores 19% on GDP.pdf, level with MiMo-V2.6-Pro at 19% and behind Kimi K3 at 22%, an 18-point improvement over Mistral Large 3.
Perspective
The available form is a Research Public Preview on Mistral's API, with weights planned for release at the end of October; every score reported here applies to that preview, not to a final weights release.
The release note says weights will be published at the end of October but gives no exact date; the cost comparison uses standard pricing and the two-week 50% discount, after which cost per task returns to $1.13.