Hermes Agent 2.0 is INSANE! artwork

Hermes Agent 2.0 is INSANE!

AI News Today | Julian Goldie Podcast

June 30, 2026

Hermes Mixture of Agents 2.0: Beat Frontier Models with a Panel of AI Agents (Agent OS Demo + Benchmarks)Julian demonstrates a new Hermes Mixture of Agents 2.
Speakers: Julian Goldie
**Julian Goldie** (0:00)
Hermes Mixture of Agents 2 might be one of the most powerful systems that I've ever tested out. So this is a new system I've set up with the new Hermes Mixture of Agents release. I'll explain exactly how it works. And I've tested it on 42 different builds, and it is absolutely mind blowing in terms of what it can do and how it performs and also how it performs versus everything else. So you can see some of the things that we've created right here with this setup. And basically what this does is it has multiple different models working together as a panel of agents that can build and automate anything that you want. And the point here is that you can create better outputs and higher levels of intelligence higher than any frontier that we have currently using a combination of agents working together instead of relying on waiting for these gated models like GPT 5.6 or for example, MIFOS to come out. And all of this stuff I've literally built today using mixture of experts and on benchmarks versus everything else, which I'll show you in a second. This is outperforming pretty much everything. So how does this work? Well, basically, mixture of experts is a system where you stack multiple models together inside Hermes and then you can be Opus GPT and it's not with a bigger model, but with a smarter system. And that's really what this is about. The models don't really matter. The system does. So what we've essentially done here is we've combined the newest mixture of agents model that came out from news research. It's just dropped on Hermes Agent yesterday. We've plugged that into the agent operating system where we can just chat with the panel. We can ask the panel anything. We can switch the APIs over here and then this aggregates whatever we create. And then we can see everything that will be built inside our workspace over here.
So if we ask it to do something like so, it will, for example, begin to run the panel. And then, after that, how it works essentially is that you have one prompt and that will run through multiple different models and it will be aggregated into one better answer. Now the other power of this, of course, is that you're using Hermes Agent, which is the most popular, powerful, open-source, free agent in the world, which means that when you combine the agentic capabilities of something like Hermes with the power of something like Mixture of Agents, you're creating something that's really, really powerful. Now you might say, okay, that's great, Julian, this stuff looks cool, but how does it perform against, for example, all the other frontier models that we have? Well, right now, I've got a leaderboard and I test, again, I've tested this out on 42 different models. I've used exactly the same prompts across every single model that's out there that's actually useful and the two top systems and the keyword is systems here, at the top of the leaderboards are Fusion and Hermes Mixture of Agents. Claude Opus is actually down here. If you're wondering why, well, I'll show you that in a second.
So I literally combined and created all this stuff and you can test it out for yourself, see for yourself what you think on Goldie Bench. But if we actually go to the compare tab here and we compare Hermes Mixture of Agents versus, for example, Claude Opus 4.8, Hermes Mixture of Agents is absolutely dominating. The stuff that you can create looks way nicer than the stuff with Opus 4.8. Let me show you, for example, this looks way nicer than this buggy output that we got using the same technique. You can't even really control properly.
Opus 4.8 is still good, but Mixture of Agents basically makes it even better.
It puts it into a better system where you can get more out of it. Also, you're using the agentic power of Hermes alongside that. The cool thing about this is, for example, we've built out these open world games, for example, like so and if we compare this against something like this, it's just not quite as nice, not quite as good.
Overall, from what I've seen, have a look for yourself if you want to compare them side by side. But from what I'm seeing on Goldie Bench, Hermes Mixture of Agents is absolutely destroying everything. This is a trend that I've seen recently, it's like Fusion and Hermes Mixture of Agents, both systems, both panel-based models and Fugu Ultra, are all right at the top of the leaderboard because they just combine the best of all models and then have them working together. That makes sense to me. If you're wondering how to use this, well, basically, you can go inside these terminal and you can run the Mixture of Agents terminal commands like so. This is technical and that's why I actually put together instead of something that I call the Hermes Agent OS, which you can see here, it's way simpler and easier to use, right, because you can basically just go inside here and go to the Mixture tab and then you can type in your prompt here, run the panel and also the cool thing is you don't need any terminal commands, you can just run this whenever you want. And additionally, when you're using this as well, the cool thing is that all of your stuff just gets saved inside this workspace. So it's easy to preview, it's easy to come back to. And also you can combine that with the chat with Hermes Agents. What we actually have inside the chat is Mixture of Agents that we can chat to as well at any point. We have the talk mode where we can talk to our agents. We have Hermes Jarvis, which we can voice control and it can use computer use, browser control, and we can preview anything that we've built with it. And then we have of course the Hermes Oracle, which you can pull in the latest news and then we can draft content using it. We have the studio, but Mixture of Agents over here. The other cool thing that I like about having a system like this is like when you're doing outreach, for example, I can just build my own SaaS tool for getting leads and building email campaigns and finding the leads in like one single click using this system. And then we can enrich our campaigns, send out emails, boom, shakalaka. It's all in one inside one place, right? Way smoother and easier to use than if you're just inside the terminal or Hermes desktop. So this is what we're finding so far. So a panel of experts beats one genius.

5 more minutes of transcript below

Feed this to your agent

Try it now — copy, paste, done:

curl -H "x-api-key: pt_demo" \
  https://spoken.md/transcripts/1000651996090

Works with Claude, ChatGPT, Cursor, and any agent that makes HTTP calls.

From $0.10 per transcript. No subscription. Credits never expire.

Using your own key:

curl -H "x-api-key: YOUR_KEY" \
  https://spoken.md/transcripts/1000774787590