NVIDIA-Accelerated Mistral 3 Open Models for Efficient and Accurate AI Performance

NVIDIA has introduced the Mistral 3 open model family, delivering industry-leading accuracy and customization capabilities. The new models include a large state-of-the-art sparse multimodal and multilingual mixture of experts (MoE) model with 675B parameters and a suite of small, dense high-performance models (Ministral 3) with sizes 3B, 8B, and 14B. The models are optimized for NVIDIA GB200 NVL72 to edge platforms and are available through Mistral AI on Hugging Face. NVIDIA-accelerated Mistral Large 3 achieves best-in-class performance on NVIDIA GB200 NVL72.