01What is Hermes Mixture of Agents?
Hermes Mixture of Agents is a way of combining multiple AI models together to get outputs that are often better than any single model. Instead of relying on one model, it uses a panel of agents that work together and a judge that fuses the answers into one final result.
This method works by giving every draft a second opinion. The judge critiques the outputs, and a chair takes the best ideas from each piece of code. This evidence-based approach helps improve quality compared to using a single frontier model alone.
In this episode, Julian Goldie tested Hermes Mixture of Agents against Claude Fable 5 and OpenRouter's Fusion. He built 42 different projects to see which one performs best in real-world tasks.
02How Hermes Mixture of Agents Works
The system uses multiple agents working together as a panel. Each agent creates a draft, then a judge fuses all answers into one final output. A chair looks at the best options from each model and plugs them together.
Here are the key parts of the system:
- Panel of agents: multiple models work together.
- Judge: critiques outputs and quality controls the final answer.
- Chair: selects the best ideas from each draft.
- Evidence-based routing: the judge looks at actual evidence to improve results.
According to Julian, combining models like GPT 5.5 and Opus 4.8 can produce outputs that are better than a newer frontier model like Opus 4.8 alone. This is backed by benchmarks and personal testing.
03Setting Up Hermes Mixture of Agents
To get started with Hermes Mixture of Agents, you need to access it through a platform that supports it. In the episode, Julian mentions that you can plug it into a system like the AI Profit Boardroom, which includes the Hermes agent.
- Go to the AI Profit Boardroom website or use the link provided.
- Find the daily update section to get the latest agent OS system.
- Download the zip file for the system.
- Follow the video tutorial to install it.
- Connect your API keys for the models you want to use.
- Start creating with Hermes Mixture of Agents.
Julian also notes that you can use the system with Claude CLI for easy back-and-forth. Everything is stored inside the workspace in your agent OS.
04Testing Hermes Mixture of Agents in Real Projects
Julian tested all three systems side by side on multiple projects. Here are the examples he showed:
- Game creation: Fable 5 won the first game test with the nicest output.
- Nordic Crypt game: Fable 5 won again for ease of navigation and detail.
- Orbit test: Hermes Mixture of Agents won with good controls and drag features.
- Landing page: Hermes Mixture of Agents produced the nicest output with glowing effects.
- Dragon Realm: All three tied, but Hermes Mixture of Agents was slightly better.
- Web OS: Fable 5 worked best; Hermes Mixture of Agents was buggy.
- Voxel: Fusion won this round.
- Frostfell: All three tied.
- Doom: Hermes Mixture of Agents and Fusion won.
In the web OS test, Hermes Mixture of Agents had a nice background but the notes app moved around on its own. The paint app worked okay, but overall it was buggy.
Finally, Julian says, 'Don't listen to the benchmarks, just test it out yourself.' This shows the importance of real-world testing.
05Results and Comparisons from the Episode
Over 42 different builds, the results were close. Julian counted the wins from the shown tests:
- Fable 5 won three rounds: the first game, Nordic Crypt, and Web OS.
- Hermes Mixture of Agents won three rounds: orbit, landing page, and Doom (tied with Fusion).
- Fusion didn't win any rounds outright as of the test.
Julian notes that Fusion claims to beat Fable 5 on benchmarks, but his tests didn't show that. In the landing page test, Hermes Mixture of Agents clearly outperformed Fable 5.
He also says that Fable 5 creates nice outputs but struggles with UI at times. Hermes Mixture of Agents often produced smoother and more polished results.
06Tips, Limits, and Mistakes to Avoid
There are some important tips and limits to keep in mind when using Hermes Mixture of Agents:
- It's not always perfect—some outputs were buggy, like the web OS and Voxel.
- Hermes Mixture of Agents can be slow for complex tasks, so it's best for 10-15 minute prompts.
- Always test yourself instead of trusting benchmarks.
- If you're on a Claude subscription, you may run out of tokens quickly with Fable 5.
Julian suggests using Fable 5 as your default because it's easier to go back and forth and works well. But when you run out of tokens, Hermes Mixture of Agents is a great backup.
He also warns that some outputs look nice on the outside but don't work well, like the web OS that didn't auto-save. Always check functionality.
07Who Should Use Hermes Mixture of Agents
Hermes Mixture of Agents is for anyone who wants better AI outputs without needing the latest frontier model. It's especially useful if you're on a budget or run out of tokens on plans like Claude's subscription.
Julian, who builds with Hermes and Claude, says you can get everything in one system with the AI Profit Boardroom. This includes an SEO agent, a video agent, and a memory system.
If you're a developer, creator, or business owner, using Hermes Mixture of Agents can give you high-quality results for games, websites, and more. It also works well as a backup when Fable 5 hits its limits.
