01What Mixture of Agents Is and How It Works

Mixture of agents is a method that combines multiple AI models into one system to get better answers than using a single model alone. Instead of relying on one powerful model, you let several models work together in parallel, then merge their outputs. This is similar to how a panel of experts can outperform one genius.

In this episode, Julian explains that the approach is a workaround for gated access to frontier models like Fable 5 or GPT 5.6. Because those models are limited, you can still achieve high-quality results by mixing available models. The process is simple: one question goes in, several models think separately, and an aggregator combines their answers into one clean output.

02Key Features of Hermes Mixture of Agents

Hermes Agent's new Mixture of Agents preset offers several useful features. It allows you to combine multiple models you already have access to. You can choose from preset options that work together in parallel, similar to other systems like Fusion or Sakana Fugu.

  • Preset configurations: Pre-made formulas for mixing models.
  • Provider agnostic: Works with any provider, not limited to specific ones like News Portal or Open Router.
  • Terminal commands: Use commands like 'model default provider MOA' to switch between presets.
  • Dashboard configuration: Configure presets directly from your dashboard or desktop app.
  • Agent loop support: Integrate mixture of agents into your agent loops for better performance.

These features make it easy to adopt mixture of agents without needing extensive technical knowledge.

03How to Set Up Mixture of Agents in Hermes

Getting mixture of agents working in Hermes is straightforward. Follow these steps to start combining models.

  1. Open your terminal and run Hermes update to ensure you have the latest version.
  2. Alternatively, go to your dashboard, navigate to the Hermes manage section, and click on Update Hermes.
  3. After updating, type Hermes model in your terminal.
  4. Look for the option labeled mixture of agents and press Enter to access it.
  5. Select a preset that suits your needs, or configure your own combination of models.
  6. Once configured, go to your model settings in the dashboard and load the mixture of agents.

You can also configure presets using the desktop app or by running Hermes MOA configure. The process is designed to be simple and quick.

04Using Mixture of Agents in Your Workflow

To get the most out of mixture of agents, integrate it into your daily AI tasks. In the episode, Julian shows how to use it within Hermes Agent's dashboard. You can chat with the system, switch between different models, and see all your built projects in one place.

  • Switch presets: Use the command 'model default provider MOA' to select different model combinations.
  • Apply to agent loops: Use mixture of agents in your loops for tasks that require higher accuracy.
  • Combine with other tools: Use mixture of agents alongside Fusion or Sakana Fugu for even more flexibility.

For example, if you're building an agent that needs to answer complex questions, you can set up a mixture of Opus and GPT. The system will generate answers from both and merge them, resulting in a more comprehensive response than either model could give alone.

05Performance Results and Comparisons from the Episode

The episode presents benchmark data showing the effectiveness of mixture of agents. According to Hermes Bench, a reference mixture of agents scores 8% higher than Opus and 11% higher than GPT. This indicates a clear improvement in performance.

Julian also compares mixture of agents to other similar systems. He shows that Fusion, which works on the same principle, outperforms most models on the GoldieBench leaderboard. The idea is that a panel of experts beats a single genius, even if each expert is less capable.

By combining cheaper models, you can achieve better outputs than using one expensive model alone.

06Tips, Limitations, and Mistakes to Avoid

While mixture of agents is powerful, there are some things to keep in mind. First, it may use more tokens because multiple models process the same query. However, the improved output often justifies the cost.

Second, in Hermes Agent, the current setup seems to support only two models per preset, so you can't combine more than two at once.

  • Don't wait for next models: Instead of waiting for the next big release, mix existing models to get better results now.
  • Start simple: Begin with preset configurations before creating custom ones.
  • Test and iterate: Try different model combinations to find what works best for your tasks.

Avoid the mistake of sticking to one model. The lesson is to build a system around your models, not just rely on a single one. This approach helps you move beyond model ceilings and avoid gated access issues.

07Who Should Use Mixture of Agents and Why It Matters

Mixture of agents is for anyone who uses AI models and wants better results without waiting for the next releases. It's especially useful for developers and businesses that rely on AI for critical tasks. The method allows you to achieve frontier-level quality with models you already have access to.

In this episode, Julian emphasizes that the model is not the moat; the system around it is. By using mixture of agents, you can squeeze more performance out of your existing models. This is a practical way to stay competitive without expensive or gated frontier models.