01What Is Nemetron 3 Ultra and How It Works

Nemetron 3 Ultra is a free, open-source AI model from NVIDIA and News Research, built specifically for long-running AI agents. It has 550 billion total parameters but uses a mixture-of-experts design, meaning it only activates the parts of the model needed for each task. This makes it faster and more efficient than traditional models.

The model is five times faster than previous versions and uses 30% fewer tokens on agentic tasks. It is designed to run for hours, handling planning, tool use, failure recovery, and decision-making without losing track of the goal. You can use it for free on News Portal for two weeks, or run it locally if you have the hardware.

02Key Features of Nemetron 3 Ultra

Nemetron 3 Ultra comes with several standout features for AI agent work. It is a frontier-grade model tuned specifically for agent setups, not just chat. Here are the main features:

  • Five times faster than previous models, making it quicker for multi-step tasks.
  • Mixture of experts: only fires the parameters needed for each step, saving compute and energy.
  • 30% fewer tokens used on agentic tasks, reducing costs and latency.
  • Open source: you can download and run it locally if you have the setup.
  • Long-horizon planning: can work on tasks for hours without losing context.
  • Post-trained for AI agent harnesses like Hermes Agent and OpenCore.

03How to Set Up Nemetron 3 Ultra for Free

Setting up Nemetron 3 Ultra is simple and takes just a few minutes. Follow these steps to get started:

  1. Sign up at News Portal for free.
  2. In your Hermes Agent dashboard, go to the manage section and then to models.
  3. Select the model NVIDIA Nemetron 3 Ultra from the list. You can now use it in chat or hand it long tasks in goal mode.

You don't need to use the terminal anymore, thanks to the new mission control dashboard. Other free models like Gemini 1.5 Flash are also available on News Portal.

04How to Use Nemetron 3 Ultra in Hermes Agent

Once Nemetron 3 Ultra is set up, you can use it in two main ways. First, you can chat with it directly to test its responses. Second, you can use goal mode in Hermes Agent.

In goal mode, you give the agent a big task like building a website or completing an SEO strategy. The agent will work on that task for up to 20 turns (or more if you set it). A judge analyzes each turn to see if the goal is complete.

If not, the agent keeps going until done.

Julian Goldie showed an example where he asked the agent to create a living galaxy. The model handled the multi-step reasoning and planning without getting stuck. This makes it great for complex tasks like coding, deep research, and long-running automation.

05Benchmark Results and Comparisons

In this episode, Julian Goldie compared Nemetron 3 Ultra to other top models using several benchmarks. Here are the results:

  • Agent productivity: Nemetron 3 Ultra outperformed GLM 5.1 and Qwen 3.5.
  • Instruction following: it outperformed all compared models.
  • Long context: it outperformed everyone tested.
  • Long horizon planning: it outperformed KimiK 2.6 and Qwen 3.5, but not GLM 5.1.

Overall, the model excels in professional work tasks, long context handling, instruction following, and agent productivity. It was post-trained for AI agent harnesses like Hermes and OpenCore, which makes it a strong choice for agentic workflows.

06Tips, Limits, and Mistakes to Avoid

When using Nemetron 3 Ultra, keep these tips in mind. First, remember that the free access on News Portal is only for two weeks. After that, you may need to switch to another model or pay.

Second, the model is designed for long-running agents, so don't use it for simple one-turn chats—it's overkill. Use it for tasks that require planning and multiple steps.

Third, make sure to use the mission control dashboard to change models instead of the terminal. It's much easier and faster. Finally, if you run the model locally, you need powerful hardware.

For most users, News Portal is the best option. Avoid thinking that free models are weak—this is a frontier model from NVIDIA.

07Who Should Use Nemetron 3 Ultra and Why

Nemetron 3 Ultra is ideal for anyone who needs a powerful AI agent that can work on complex, long-running tasks. This includes developers building AI agents, businesses automating workflows, and researchers doing deep analysis. It's also great for people who want to test a frontier model without paying, thanks to the two-week free offer.

If you use Hermes Agent or OpenCore, this model slots in easily. It saves time because it can work for hours without your input. Julian Goldie's community, the AI Profitable Audium, also offers tutorials and support for getting the most out of the model.

Whether you're a beginner or expert, the model's open-source nature and long-horizon capabilities make it a valuable tool for AI automation.