Live data from Hacker News

Nemotron 3 Ultra: Open Moe Hybrid Mamba-Transformer for Agentic Reasoning [pdf]

research.nvidia.com

Post reply on HN

Print view View original

1–3 of 3 posts

Nemotron 3 Ultra: Open Moe Hybrid Mamba-Transformer for Agentic Reasoning [pdf]

#1

Post by victormustar » Thu, Jun 04, 2026, 1:06 PM UTC

Nemotron 3 Ultra: Open Moe Hybrid Mamba-Transformer for Agentic Reasoning [pdf]
research.nvidia.com

Re: Nemotron 3 Ultra: Open Moe Hybrid Mamba-Transformer for Agentic Reasoning [pdf]

#2

Post by throwa356262 » Thu, Jun 04, 2026, 2:34 PM UTC

Is this the one from Jensens Computex presentation the other day?

It is significantly bigger than Qwen for the same level of intelligence, but I think the key strength was inference speed.

Re: Nemotron 3 Ultra: Open Moe Hybrid Mamba-Transformer for Agentic Reasoning [pdf]

#3

Post by 2001zhaozhao » Thu, Jun 04, 2026, 6:07 PM UTC

This model seems like a really big deal. Is this the biggest Western open-source AI model in the world (beating out Llama3 405B)?

Post reply on HN