Mistral AI introduced the Mistral 3 family of open-source models optimized for NVIDIA supercomputing and edge platforms. Mistral Large 3, a MoE model, offers efficiency, accuracy, and scalability with 41B active parameters and 675B total parameters, available on Dec. 2. The collaboration with NVIDIA enables distributed intelligence and improved performance.

This partnership combines NVIDIA GB200 NVL72 systems with Mistral AI’s MoE architecture to deploy massive AI models efficiently. Mistral Large 3 achieved performance gains and energy efficiency on the GB200 NVL72 compared to previous models, setting the stage for distributed intelligence. The compact Ministral 3 suite is optimized for NVIDIA edge platforms.

Developers can access the Ministral 3 suite via Llama.cpp and Ollama for efficient AI on the edge. Mistral AI’s models, openly available, empower customization and acceleration of AI innovation. By linking with NVIDIA NeMo tools, enterprises can customize models for their use cases and optimize inference frameworks for efficiency from cloud to edge.

The Mistral 3 models are available on open-source platforms and cloud service providers, soon deployable as NVIDIA NIM microservices. These models are set to revolutionize AI deployment across industries. The collaboration aims to democratize access to frontier-class technologies and accelerate the transition from prototype to production.

Read more at Nvidia: NVIDIA Partners With Mistral AI to Accelerate New Family of Open Models