NEW Qwen 3.8 Max + Hermes Agent Just Changed Everything

**Julian Goldie** (0:00)
Qwen 3.8 Max just dropped. I've already plugged it into Hermes Agent, and it's pretty powerful from what I've tested so far. So what I'm going to show you today is how to use Qwen 3.8 Max with Hermes Agent, and also how to get the most out of it, and some of the cool stuff that we've created with it so far. So the great thing about Qwen 3.8 Max is that it's basically Fable 5 level, and you can use your coding plan with it to plug it in directly into Hermes, which you actually can't do with Fable 5 unless you use the API. It's also a 2.4 trillion parameter brain, and it actually beats Fable 5 on some of the key benchmarks. So basically, Hermes agent is the hands, and then you can change the brain. So you can plug in Qwen 3.8 Max. It's also 2.4 trillion parameters, one million context window, and also designed for agentic tasks. I'll show you some things we've built with it so far. So it created like some pretty cool coding apps, as you can see right here. But also, we actually linked it to MCP. So we linked it to Higgs Field MCP, just to test it out and see what it could do. And we actually used and gave it our book image as an example to see if it could code and build some cool stuff with Higgs Field MCP and Qwen 3.8 Max inside Hermes agent. So this is some of the stuff that it created, which is pretty nice, and it was super easy to use. Another thing that would say about Qwen 3.8 Max is pretty fast to respond. Here's another example. So what we actually have inside our Hermes agent operating system is all of our different profiles with different models plugged in. Now, one thing I'll say about Qwen 3.8 Max is it's very fast to reply. So we say, for example, learn, and then give it access to a guide. And you can see here it says skill learned and registered. Here's a skill. Here's the path that is saved locally as a skill MD file, which we can open up in a second. And here's what it captures. And so we've taught it a new skill in the space of a couple of minutes. And if we open this up, we've got the full skill that's written here based on the guide we actually trained it with. Now, one thing I definitely recommend with this is that when you're using new models, whether that's Fable 5 or whether that's Qwen 3.8 Max, whatever comes out next, you can actually plug in a memory system here so that has context. And the great thing about this is if we use, for example, Hermes Agent OS like this, you can see that actually saves the skill inside the Obsidian memory. Why is that useful? It's useful because, for example, if we're switching between the memory system inside Hermes, and then we go over to and we use, let's say DeepSeek, where it's got that memory stored from Hermes Agent OS. The same, for example, if we go over to Claude Code over here, and we select Qwen 3.8 as the brain, we can now use Qwen 3.8 as the brain of Claude Code, but then also switch over to Hermes Agent OS as well whenever we want to.
And so you can describe what you want.
The 2.4 trillion parameter brain thinks it through, Hermes creates whatever you ask it to do, and then you can actually open that up in your browser. Also, some people say like, this is difficult to set up a new model inside Hermes Agent OS. So you can literally just go to the Manage tab or your dashboard with Hermes Agent. I would make sure that you update first, so you can just click the Update button. And then from here, you can create a new profile with Qwen 3.8 Max plot in.
So you can see here that we have a separate profile for Qwen 3.8 ready to go.
Now, also, when they were training Qwen 3.8 Max, this model is very good at coding and actually ran 10 days autonomously with self-evolving development from an empty folder to production without handholding. So basically, it updated and you can trace what it's done inside the GitHub for this model and see exactly what it built. And so it can code autonomously for hours. Now, why is that important with Hermes agent? Well, essentially, if you're using the goal mode, and we have a separate section for goal mode over here, we can give it a goal and then it can go off and complete the goals and run autonomously for hours without us even having to touch it. So we can set the target and then just walk away with goal mode. And Qwen 3.8 Max is designed to actually do that. The other benefit of this, and this is something I haven't tried yet, but would be super interesting, is you could use Hermes Mixer of Agents and you could use, for example, Fable 5 combined with Qwen 3.8 Max. And that fuses the answers together so you can get even better outputs than if you were coding with Qwen 3.8 Max alone inside Hermes.