AI BriefingNVIDIA Announces Launch of Nemotron 3 Ultra
Key point
NVIDIA has unveiled Nemotron 3 Ultra, the most powerful US open-weight model at 550B parameters.
Details
NVIDIA announced Nemotron 3 Ultra during its Computex keynote. With a total of 550B parameters (55B active parameters), it is the largest model in the Nemotron 3 series and currently the most intelligent open-weight model released in the US.
On the Artificial Analysis Intelligence Index, it scored 48 points, surpassing major competing models such as Gemma 4 31B (39 points) and gpt-oss-120b (33 points). However, it still trails somewhat behind China's Kimi K2.6 (54 points).
In terms of speed, it also shows standout performance. On DeepInfra endpoints, it processes over 300 tokens per second, significantly exceeding the speed of China-based models like DeepSeek and Moonshot (Kimi), which run at 50-100 tps.
The model features approximately 90% sparsity, and will be offered in BF16 weights as well as an NVFP4 quantized version for higher inference performance.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Stanford and NVIDIA Release 'Contrastive Language Model' CLM with 9x Faster InferenceInception Releases Ultra-Fast LLM 'Mercury 2.5' Generating 770 Tokens Per Second