Nvidia has launched Nemotron 3 Ultra, its most advanced open-weight AI model to date, marking a major step forward for the United States' open AI ecosystem. Yet despite the breakthrough, Chinese developers continue to hold the lead in the increasingly competitive race for open-weight artificial intelligence.

Unveiled by Nvidia CEO Jensen Huang at Computex 2026 in Taipei, the model combines faster performance, lower operating costs and stronger benchmark results than previous generations. However, it still trails China's top-performing open-weight models, highlighting how rapidly Chinese AI labs have advanced over the past two years.

Nvidia's Biggest Open AI Leap Yet

Nemotron 3 Ultra is built on a mixture-of-experts architecture with 550 billion total parameters, though only 55 billion are activated at any given time. The design allows the model to allocate computing resources more efficiently, reducing costs while maintaining high performance.

According to Nvidia, the model can generate more than 300 tokens per second, deliver inference speeds up to five times faster than some competing systems and lower operating costs by roughly 30%.

The company believes those advantages could make Nemotron 3 Ultra particularly attractive for businesses developing AI agents, automation tools and other computationally intensive applications.

Independent testing by Artificial Analysis awarded Nemotron 3 Ultra an intelligence score of 48, based on factors including reasoning, coding, general knowledge and agent performance.

The result places the model ahead of other US open-weight competitors, including offerings from Google and OpenAI. It also represents a significant improvement over Nvidia's previous Nemotron 3 Super model, which scored 12 points lower on the same benchmark.

In a sector where incremental gains are becoming increasingly difficult to achieve, the jump underscores Nvidia's growing ambitions in the open AI market.

China Remains the Benchmark

Despite Nvidia's progress, the overall lead remains with China. Moonshot AI's Kimi K2.6 scored 54 on the same Artificial Analysis benchmark, maintaining a noticeable advantage over Nemotron 3 Ultra. The model is widely regarded as one of the strongest AI systems currently available, regardless of whether it is open-weight or proprietary.

The gap reflects a broader shift in the global AI landscape. While many US technology firms continue to reserve their most advanced models for closed commercial platforms, Chinese developers have increasingly embraced open-weight releases, allowing them to gain influence across the global AI ecosystem.

Nvidia is betting heavily that open-weight AI will remain strategically important. The company has committed to a $26 billion development plan over the next five years and is already developing Nemotron 4 alongside partners including Mistral AI and Perplexity through the Nemotron Coalition.

Nemotron 3 Ultra demonstrates that Nvidia is emerging as America's most ambitious open-weight AI contender. But the launch also serves as a reminder that the competitive landscape has changed.

For years, the world's most advanced AI models largely came from the United States. Today, Chinese labs are increasingly setting the pace in open-weight AI, forcing American companies to accelerate their efforts in a race that is becoming as much geopolitical as technological.