Hermes just dropped one of its most useful updates yet: Hermes Mixture of Agents, or MoA. I’ve been running this exact pattern for weeks, so let me show you what it is and why it matters.
In short, Mixture of Agents lets several AI models work together in parallel and combine into one stronger answer — a panel of experts instead of a single model. It’s how you get frontier-level quality without waiting for gated models.
Key takeaways
- Hermes Mixture of Agents runs multiple models in parallel and aggregates them into one stronger answer.
- It’s a smart workaround for gated, preview-only frontier models like Fable 5 and GPT-5.6.
- On Hermes Bench, a two-model panel (Opus 4.8 + GPT-5.5) beats either model alone — and it’s one command to switch on.
What Is Hermes Mixture Of Agents?
Mixture of Agents is a virtual model provider inside Hermes. You pick a named MoA preset, and instead of one model answering, several models each give their take privately, then an aggregator model reads them all and writes the final answer.
Think of it like a panel: one brilliant person answering a hard question alone versus a panel of experts each writing their own take, with a sharp chair combining the best of all of them. The panel wins almost every time. That’s exactly what MoA does for your AI agent.
Why This Matters Right Now
Here’s the timing. The newest frontier models are getting gated — Claude Fable 5 is only rolling out to a handful of approved partners, and GPT-5.6 is a limited preview. If you want frontier-level intelligence, you often simply can’t get access.
Mixture of Agents is the workaround. Instead of begging for access to one locked model, you combine several models you already have into something that punches above any of them. You go beyond the public frontier with no extra access needed.
Panel Beats Genius: The Benchmark Numbers
Does it actually work? According to Hermes Bench, yes. A two-model MoA preset — Opus 4.8 aggregating over a GPT-5.5 reference — outscores either model on its own:
- Opus aggregator (Opus 4.8 + GPT-5.5) — MoA: 0.8202
- Claude Opus 4.8 alone: 0.7607
- GPT-5.5 alone: 0.7412
That’s the panel beating its strongest single member by around 6 points — and Hermes reports the reference combo scoring roughly 8% higher than Opus and 11% higher than GPT. Combining perspectives genuinely lifts quality on hard tasks, rather than just averaging them out.
How To Set Up Mixture Of Agents
The setup is refreshingly simple. Here’s the flow:
- Run
hermes updatefirst (terminal or the dashboard’s Manage section) - Run
hermes modeland pick the Mixture of Agents provider - Choose a named preset, or configure your own
- Switch anytime with
/model default --provider moa, or use the/moashortcut for a one-off
It’s provider-agnostic, so you can plug in whatever models you like. You can even configure presets in config.yaml with explicit reference models and an aggregator.
The Best Combo To Start With
If you just want the strongest setup out of the box, the top performer on Hermes Bench is an Opus 4.8 aggregator with a GPT-5.5 reference. That single preset beats either model running alone.
The clever part: you can even use cheaper models together and still get better outputs than one expensive model in isolation — frontier quality for a fraction of the cost.
Stop Chasing The Model, Build The System
Here’s the bigger lesson I keep coming back to. Everyone is waiting on the next model — the next Opus, the next GPT — to finally change everything. But a mix of today’s models already beats the best single model that you can’t even access.
The model is the part you swap; the system is the thing you own. Stop chasing the model and build the system instead. Mixture of Agents hands you that lesson for free.
Where I Run All Of This
I run Mixture of Agents inside my Agent OS, alongside two other systems built on the same idea — Fusion and Sakana Fugu — all wired into one dashboard. One click to switch between them, with everything I’ve built ready to preview in the same place.
If you want that whole stack done for you — the Agent OS zip, MoA, Fusion and Sakana presets, plus live coaching where I build model panels with you — it’s all inside my AI Profit Boardroom, with 3,800+ operators running the exact same setup. New to Hermes? Start free with my AI Money Lab, and see my guide to the best models for Hermes agent.
Mixture Of Agents vs Fusion vs Sakana Fugu
If you follow my channel, you’ll know Mixture of Agents isn’t the only system built on this panel-of-models idea. Fusion and Sakana Fugu do something similar — multiple models combining to reach near-frontier intelligence.
You don’t have to pick just one. In my Agent OS I keep all three wired in and switch between them with a click, depending on the task. They’re three takes on the same powerful principle: a panel of today’s models beats waiting for tomorrow’s gated one.
FAQ
What is Hermes Mixture of Agents?
It’s a feature that runs several AI models in parallel and aggregates their answers into one stronger response — a panel of experts instead of a single model.
Why use it instead of one model?
Because gated, preview-only frontier models are hard to access. A panel of models you already have can beat any single one, with no special access needed.
Does it really beat a single model?
On Hermes Bench, an Opus 4.8 + GPT-5.5 panel scored 0.82 versus 0.76 for Opus alone — so yes, it beats its strongest member.
How do I turn it on?
Run hermes update, then hermes model, pick the Mixture of Agents provider and a preset. You can switch with one command.
Does it cost more?
It uses more tokens because of the extra reference calls, but you can mix cheaper models and still beat one expensive model alone.
The Bottom Line
Hermes Mixture of Agents is one of those updates that quietly changes the game: frontier-level quality from models you already have, with one command to switch it on.
Stop waiting for the next gated model. Build a panel, own the system, and let it outperform the genius working alone — that’s the whole point.
