Nvidia Releases Its Best Open AI Model Yet—But Still Lags Behind China - Decrypt

Nvidia has just unveiled Nemotron 3 Ultra, its most powerful open-source large language model, alongside the Cosmos 3 platform aimed at developing advanced "physical AI" and "world models" for robotics. While Nemotron 3 Ultra now ranks as the smartest open US model, benchmarks indicate it still trails behind leading Chinese open AI models in certain complex reasoning tasks and multimodal capabilities, underscoring the escalating US-China tech rivalry. This strategic release highlights Nvidia ambitious push to integrate its dominant AI hardware with sophisticated software, positioning itself as the foundational layer for the next generation of AI, particularly in embodied intelligence. The launch, supported by the "Cosmos Coalition" with partners like AI video giant Runway, aims to accelerate the development of systems that can understand and interact with the physical world, leveraging Nemotron 3 Ultra advanced reasoning to power Cosmos 3 robotic applications. The competitive landscape remains intense as China continues to pour resources into open-source AI, with its models demonstrating rapid advancements and widespread adoption. Observers will be watching closely to see if Nemotron 3 Ultra iterative improvements and the broader Cosmos ecosystem can genuinely close the performance gap and solidify Western leadership in the burgeoning field of physical AI, or if China's momentum will continue to widen the lead in open-source innovation.