Nvidia's Nemotron 3 Ultra Becomes the Smartest Open US Model, but China Still Leads

Why It Matters

Nemotron 3 Ultra raises the performance ceiling for open‑source AI in the US, narrowing the gap with China’s leading models and pressuring closed‑source incumbents. Its speed and accessibility could accelerate developer adoption and diversify the AI ecosystem.

Key Takeaways

• Nemotron 3 Ultra has ~550 B parameters, 55 B active
• Scores 48 on Artificial Analysis, beating all open US models
• Delivers >300 tokens/second on DeepInfra, outpacing peers
• Launch scheduled for June 4 across Hugging Face and OpenRouter
• Chinese Kimi K2.6 still leads open‑weight rankings at 54 points

Pulse Analysis

Nemotron 3 Ultra represents a quantum leap for U.S. open‑weight AI, combining a massive 550‑billion‑parameter architecture with a dynamic active‑parameter count of roughly 55 billion. This design enables the model to deliver more than 300 tokens per second on the DeepInfra platform, a speed advantage that dwarfs peers like DeepSeek and Moonshot, which hover between 50 and 100 tokens per second. By releasing the model on June 4 across major hubs such as Hugging Face and OpenRouter, Nvidia is positioning the model for rapid community integration and iterative improvement.

Despite its impressive scores, Nemotron 3 Ultra trails China’s Kimi K2.6, which registers 54 points on the Artificial Analysis index, and remains below the closed‑source benchmark champion Opus 4.8 at 61 points. This performance gap underscores the persistent advantage of proprietary training data and optimization pipelines in the closed‑AI arena. Nonetheless, the open‑source model’s 48‑point rating sets a new benchmark for U.S. offerings, signaling that the nation’s research community can now produce models that compete more closely with the best global alternatives.

The broader implications for Nvidia are twofold. First, the company reinforces its role as a hardware and software catalyst, providing the compute backbone that makes such large models feasible. Second, by championing an open‑weight release, Nvidia may stimulate a wave of developer‑driven innovation, fostering niche applications and custom fine‑tuning that closed models typically restrict. As enterprises weigh cost, transparency, and control, Nemotron 3 Ultra could become a preferred foundation model, reshaping the competitive dynamics of the AI market.