On December 2, 2025 (local time), French AI startup Mistral AI formally introduced its Mistral 3 model family. This suite encompasses three compact dense models (14B, 8B, and 3B parameters) alongside the most powerful iteration yet of Mistral Large 3. On the same day, NVIDIA announced a strategic partnership with Mistral AI, leveraging chip technology optimization to achieve a significant performance leap for Mistral Large 3 on the GB200 NVL72 system. This enhancement represents a tenfold improvement over the previous-generation H200 chip, enabling the processing of over 5 million tokens per second per megawatt of energy consumed—a remarkable feat in energy-efficient AI computing. Furthermore, Mistral AI unveiled the Ministral 3 series, a lineup of lightweight models optimized for NVIDIA’s edge computing platforms. These models are designed to operate seamlessly on devices such as RTX-powered PCs and laptops, expanding the accessibility of advanced AI capabilities.