NVIDIA Nemotron 3 Ultra Powers Faster, Cheaper Reasoning for AI Agents

NVIDIANVIDIA

NVIDIA has released Nemotron 3 Ultra, an open model designed for long-running AI agents, and provided a tutorial for its setup and demonstrations. This model aims to make complex, multi-step agentic workflows faster and more cost-effective by delivering high throughput and efficient reasoning.

NVIDIA has released Nemotron 3 Ultra, an open 550B-parameter Mixture-of-Experts (MoE) model with 55B active parameters, optimized for agent orchestration and reasoning in long-running AI agent workflows. It is post-trained for agent harnesses, enabling multi-turn planning, tool use, and error recovery, integrating Hybrid Mamba-Transformer layers for efficient long-context handling.
Total Parameters
550B
Active Parameters
55B
Inference Throughput
Up to 5x higher
Cost Reduction for Agentic Tasks
Up to 30%
Licensing
OpenMDW-1.1

This model addresses challenges in multi-agent systems where token counts and costs grow quickly. Nemotron 3 Ultra delivers up to 5x higher throughput and can lower the cost for agentic tasks by up to 30% compared to other open models, making autonomous workflows more efficient.

Nemotron 3 Ultra ships with open weights, data, and recipes under the OpenMDW-1.1 license, the work of the NVIDIA Nemotron Coalition. It's already on Perplexity Pro, OpenRouter, and Hugging Face, and plugs into agent frameworks like Hermes Agent and OpenCode.

View the full update on developer.nvidia.com
NVIDIA AI
NVIDIA AI
@NVIDIAAI
X

Nemotron 3 Ultra is here, and we've got just the tutorial to get you going. Here's how to set up Ultra in your favorite agentic harness + some great demos of the model's capabilities 👇 https://t.co/jylnpBMyj3

17retweets122likes
View on X

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Share this update