Hermes Agent + Qwen 3.7 Max is INSANE 🤯

Ajouté : 2026-05-27

[00:00:00]Hermes Agent just officially added Quen 3.7 Max. And if you use Hermes Agent or you've been thinking about it, this matters because the model powering Hermes can now be used as Quen 3.7 Max.

[00:00:12]It's one of the most powerful agentic models you can plug in. And there's also tons of other benefits to doing it, too.

[00:00:17]So, let me walk you through what actually happened, what Quen 3.7 Max actually is, and what this combination means if you're running AI agents inside your business right now. Hello, Quen 3.7 Max just shipped and Hermes agent merged in the same week. Bear in mind that Hermes is now the number one app using Quen 3.7 Max on open router and it's got a million token context window and basically when you plug in the brain of Quen 3.7 Max into Hermes, you get a much more powerful, much more capable AI agent. So you can see for example here we've actually plugged in Quen 3.7 Max.

[00:00:46]You might be wondering how to do that.

[00:00:47]So all you do is you go into your terminal like so. You'll go to Hermes model and then from here you're going to select open router. Then you hit enter once you've added an API key. And from there you can select the model. Right now make sure before you've done this you actually run Hermes update inside your terminal. That will update to the latest version. So you can actually find Quen 3.7 Max inside the model list.

[00:01:09]Otherwise when I checked it myself earlier I couldn't find it there. Bear in mind as well you can get another free model. So if you're like I don't want to use Quen 3.7 Max. I don't want to pay for a model. No problem. You can actually get our alpha for free on Open Routter and plug that into her mutation instead. And if you're wondering, okay, what is Quen 3.7 Max? So, this actually comes from the Quen team, which is Alibaba's team. It's a Chinese model and it's really good at long horizon tasks, right? So, what that means essentially is like if you're using Hermes agent and you plug this in, it can work on task for hours without you having to supervise it. Now, you might be wondering, okay, how many hours? Like, that sounds crazy. So, this is built specifically for extremely long horizons. So, in the actual demonstrations of Quen 3.7 Max, they actually set up a 35h hour fully autonomous AI agent optimization process and it went off from there. If you're wondering how to get access to it, so you can get it from Alibaba Cloud Model Studio. I don't really use that, so I just use Open Router. It's much easier to plug in and you can see the benchmarks right here. Now, one of the benefits of this is like you can go to goal mode, for example, inside Hermes and you could type forward/goal, set a target, walk away, and then it has 50 different turns at achieving that goal.

[00:02:14]So what that means essentially is that you can close your laptop, come back, Hermes has worked on that goal for hours and hours and hours. Now depending on which model you plug into Hermes, sometimes this is good, sometimes this isn't so useful. But if you have something like Quen 3.7 Max, which is designed for long horizon tasks, well, when you're building this out, it's going to be a lot more powerful. Now, how does it perform on the benchmarks?

[00:02:35]So in the test, I've actually compared it to Opus 4.6 and Quen 3.7 Max beats Opus 4.6 Max on, for example, like Terminal Bench. Now, for me personally, do I pay much attention to benchmarks?

[00:02:46]Not particularly, but if you want the headline, basically Quen 3.7 on benchmarks outperforms Opus 4.6. One thing that's worth paying attention to is the actual general agent benchmarks.

[00:02:57]This is a bit more useful, especially if you're using AI agents like Hermes and OpenClaw, because what this tells you is basically is a very agentic model, but it doesn't work quite as well as Opus 4.6 inside your agents. However, Opus 4.6 six as an API pretty expensive right whereas for example something like quen 3.7 max is a lot more reasonable now the number one app that's being used on open router with quen 3.7 max right now is Hermes agent and that's the head of open claw kilo code and claw code so a lot of people are finding this model pretty useful and that's why they're using it so much if we actually test it out so we've got her plugged into quen 3.7 max over here let's test it out and we'll just ask a simple question and then see how long it takes to reply and it comes back within a few seconds. Not bad. Now, one of the useful things with this is like it's going to work pretty well with your Obsidian Vault because it's a million token context window. So, if you've, for example, linked it to Obsidian like I have with all your knowledge base and your second brain basically set up, basically what this can do is pull in your notes from Obsidian directly into your AI agent.

[00:04:00]So, for example, every time that I use Hermes with Quen 3.7 Max, because it's got that million token context window, it can take like a whole year of notes or thousands of notes and turn that into something useful, right? So, for example, if I say, okay, use Obsidian, come up with some ideas for automations I can build today with Hermes, right?

[00:04:18]It'll actually take the notes from my Obsidian vault and then use that to pull out that information, right? So, this is basically like a full knowledge map. I didn't create any of it. My agents just basically pull this in and then organize it beautifully like so. I call it the infinite context engine. And then everyone in my life, everything that I work with, even the systems that we use, the hobbies I have, how I create social media content, it's all detailed into this map. So that when we're using it with Hermes, we get something way more useful, right? And so your context gets better. The other cool thing about this as well is when Hermes actually comes up with ideas, it will automatically add that into the obsidian bowl. So they kind of sync both ways in this positive feedback loop. And then finally, of course, there's more of a loop on this, right? So you get a 35h hour autonomous run with this if you wanted to. I'm 99% of people actually going to do that. I don't think so. But it's a pretty cool idea. I like the vibe of that. And you can see here, for example, it's come up with three different ways that we could set up right now, right? And it can actually because it's synced to Obsidian, it can use Obsidian to create that automation. And you see how it's basically created the note. So, it actually creates a note of Hermes automation ideas because Quen 3.7 Max is a gentic. It uses tools better, which means it automatically went off and did that, right? And then you can see here that it came up with the quick wins. And these are all ideas specifically around even like team members that we have in the group, right? What I work on daily, who I work with, all the SEO work. It took that into account and then started using that to come up with ideas. So, it's pretty good. It's pretty powerful.

[00:05:45]Took about like 40 50 seconds to actually come up with the ideas that it gave us. and then it broke it down into a recommendation here. The other thing that I like here is like it kind of formats it nicely as well. So, it is a pretty good model for using with this.

[00:05:57]And again, it's super easy and simple to set up with Hermes agent, right? Cuz you just go to Hermes model and Hermes update inside your terminal. Now, if you actually look at the benchmarks here, I mean, not every benchmark matters for an agent operator, but here's like a short list that does, right? So, if you look at this, this is Opus 4.6 with the benchmarks. This is Quen 3.7 Max on Terminal Bench, on S. Pro on S sw multilingual on GPQA diamond. Quen 3.7 Max is outperforming Opus. What that means in plain English is like you get the best terminal coding agent. So if you're creating a lot of coding tasks, building out apps, websites, landing pages, this could be pretty good, right?

[00:06:33]You also get the best MCP tool agent, which is very underrated, but basically MCP is a connector layer between Hermes and the rest of your tools, right? So for example, we have an MCP between Obsidian with our memory and Hermes agent. They're going to connect better because we have a better agent and a better brain for controlling the MCPS, right? And if it's better here, you get better tool calling. Additionally, you get stronger reasoning. So, this is using GPQA diamond. You got better benchmarks on HOE, HMMT, Apex.

[00:07:03]You might wonder what they are. These are just technical ways of seeing like okay, what's the best for reasoning? And for goal mode inside Hermes agent, you know, for example, we've said like build out a beautiful website about OpenCore and then it creates all of this system right here with all the files ready to go on tasks where the agent has to like plan, replan, recover. This is really what separates it between like getting stuck somewhere and actually going off and being autonomous. And then finally, you got like best long context recall, right? So the million token context window helps, but only if the model can actually find what you put in it. And because of this benchmark, it means Quen 3.7 Max actually can do that. Now again, some people want to say benchmarks easy to game. I agree with you. That's a valid concern. Does this actually work in practice? So Quen actually trained for cross harness generalization. What does that mean? It means basically it can work across multiple agent harnesses and that's why it performs so well for example on like Quen claw bench, co-work bench and skills bench. And these are basically like agent harness tests to check if the model actually works agentically. Uh the other crazy thing about this as well is like if you look at for example Anthropic with claude 4.7 the costs are pretty crazy whereas it's a lot more reasonable with Quen 3.7 Max and also Hermes is like the number one app using this right so the good thing about this is like you're not being a beta tester for the model. A lot of people have tested it and you get something that's actually been tested by a lot of people before you plug it into Hermes. I had some pretty cool ideas for like the 35h hour autonomous run. I'm not going to run that myself but there's some interesting stuff you could build with this as well. And basically what this means and why it matters is basically Quen's own report documents like a 35h hour fully autonomous kernel optimization run. Now other Frontier models on the same task for example like GLM 5.1 reached 7.3x Kim 2.6 reached 5x and Deepc V4 Pro reached 3.3x the previous version of Quen 3.6 plus reached 1.1x. But Quen 3.0 Max hit a 10x geometric on the kernel test. So like for example on a Sunday night you could like open up Hermes goals. You say okay build a complete SEO website about insert your topic including the blog individual blog post schema markup internal links ready to deploy website hit start close your laptop open Monday morning and then preview it right something like that could be absolutely amazing. So thanks so much for watching.

[00:09:17]If you want to get my full agent set up with Hermes, the goal mode, the workspace, the control room, we even created a studio for generating images, videos, and text to speech using Hermes agent inside our agent operating system.

[00:09:30]And we plug in all our AI agents. You can get that inside the AI profit boardroom. So you can have Quen 3.7 Max as the engine, Hermes as the agent setup, and then agent operating system brings it all together in a place where you can easily manage it. And this basically comes with a zip file that you can download. We have a video tutorial working you through it. We also update this regularly for you. We give you all the prompts to install it. And you can get the zip file down here. You also get access to an awesome community. I create daily video tutorials based on your questions and give you awesome answers.

[00:09:59]So for example, like Tony asked this question about agent operating system. I actually created a video for him and a breakdown here. You can jump on weekly coaching calls where you can ask questions, get help and support in real time, share your screen, meet other cool people doing the same thing. You get access to all of my best trainings on AI and AI automation and Hermes agent.

[00:10:14]Inside the map, you can connect with people in your local area and city who are using AI agents just like you. And this is all inside the AI profer border.

[00:10:20]Feel free to get it. Link in the comments description or go to the aiproferborn.com. Eyes watching.