Mistral Artificial Intelligence yesterday announced the Mistral 3 family of open-source multilingual, multimodal models and said the new models are optimized across NVIDIA supercomputing and edge platforms. The company highlighted Mistral Large 3 as a mixture-of-experts model that activates only parts of the model with the most impact instead of firing up every neuron for every token. According to the announcement, that targeted activation delivers efficiency that allows scale without waste and accuracy without compromise, positioning enterprise artificial intelligence as practical for real-world use.
Mistral Large 3 is described with exact capacity figures of 41B active parameters, 675B total parameters and a large 256K context window. The models will be available everywhere, from the cloud to the data center to the edge, starting Tuesday, Dec. 2. The release frames the combination of Mistral Artificial Intelligence architecture with NVIDIA hardware as a route to deploy and scale massive models more efficiently by leveraging advanced parallelism and hardware optimizations built into NVIDIA GB200 NVL72 systems.
The companies present the collaboration as a step toward what Mistral Artificial Intelligence calls distributed intelligence, aiming to bridge research breakthroughs and practical applications. The announcement emphasizes enterprise-focused accuracy and efficiency, and the partnership centers on making the Mistral 3 family broadly deployable on both supercomputing and edge infrastructure. Technical details and deployment guidance are referenced in the partner announcement and the linked NVIDIA developer blog about NVIDIA-accelerated Mistral 3 open models.
