← Back to search

China's NEW Xiaomi MiMo V2.6 AI Is FREE!

AI News Today | Julian Goldie Podcast · 2026-09-23 · 8 min
relevance 86 1588 words Episode page ↗ Audio ↗
✨ Episode Outline — click any point to jump to it in the episode
Problem solved
How to get frontier-level performance for free using Xiaomi MiMo V2.6 Pro and Flash via OpenCode and Hermes Agent instead of paying frontier API prices.
Benefits
  • Free one-week access to MiMo V2.6 Flash and V2.5 inside OpenCode
  • Open-source omni-modal models with computer use
  • Pro roughly on par with Claude Opus 5 and GPT 5.6 Sol on most benchmarks
  • Pluggable as a brain in Hermes Agent profiles as a free backup to Claude
  • Cheap API: $0.87/M output tokens Pro, $0.28 Flash
Use cases
  • Plugged MiMo Pro and Flash into Hermes Agent as separate profiles for free-model jobs
  • Built dashboards, a website design, and a 'warp' demo inside the Agent OS workspace
  • Built a 3D game demo with Pro; decent builds also from Flash
  • MiMo Code UI switching between Pro/Flash and Quick/Think/Deep Thinking modes
  • Selecting Claude Opus via new Hermes Agent Claude plugin dropdown in Agent OS
KPIs / results
  • Pro: 1T total params, 426B active, 1M context (200K on OpenCode free); Flash: 309B/15B active
  • Artificial Intelligence Index 46; Terminal Bench 89.9 (best in category)
  • Deep 71.9 vs Opus 5 74; Terminal Bench 4.0 34.9 vs Opus 49
  • Pro Ultra Speed up to 20x faster output; five releases in 17 months
Tools / build
  • Xiaomi MiMo V2.6 Pro / Flash
  • OpenCode
  • Hermes Agent MiMo profiles
  • MiMo Code
  • Agent OS with Claude plugin
0:00 / 0:00
Mi just released Xiaomi MiMo V2.6 Pro and Flash. These are frontier level open source AI agents. So you've got two omni-modal models advanced through scaled reinforcement learning. Pro performs on par with Claude Opus 5 and GPT 5.6 Sol across most benchmarks. And Pro actually scores 46 on the Artificial Intelligence Index, which is pretty decent. And stronger coding as well. Computer use available. It's also open model way. It says open and available and open source. That doesn't mean you host it, but it does mean that it's open source. So just bear that in mind, like it'd be a massive model to host. And also you can use it for free. So you can actually use it for free one week inside OpenCode. So if you go to OpenCode, just make sure that you've updated it. And then once you've done that, you can select MiMo V2.6. Just to be clear here, you can see MiMo V2.6 Flash Free. So you can use it for free along with MiMo V2.5 Free as well. It's pretty nice. So you can get access to Frontier model for free. And this is a new Chinese open source model as well. So if you're looking at the power of this as well and the benchmarks, you can see how it performs. We've already plugged it into our agent OS and built some core stuff with it as well. And it's pretty easy and simple to use too. So if we go to MiMo Code over here, you can see that we can switch between Pro and Flash, Quick, Think and Deep Thinking. And we can use this system for free. Now, if you're wondering, okay, how does it perform? Well, you can actually see that we've got a bunch of stuff inside our workspace over here too. All right. So you can see some of the dashboards that we've built, some of the cool stuff. It's pretty nice for website design, as you can see right here. We've created a nice little design there. And then also, for example, you can see this kind of crazy warp thing that we built here as well. So the cool thing about this is you can use it inside an agentic operating system alongside all your AI agents. You can actually use the brain for free and then plug it into its agent as well. It's open source and it's Frontier level too. So how do these models compare? What's the difference between them? Well, MiMo V2.6 Pro is their most powerful model. One trillion total parameters, 426 billion parameters active per token and one million token context window. I did notice that if you're using it with open code, the context window is actually a lot lower, but it's a free version. So you can see here 200,000 context window on there. And then also you can see MiMo V2.6 Flash over here. 309 billion parameters, 15 billion active per token and one million token context window. And it can build some fun stuff like you can see right here. This was actually built with MiMo Flash. So just bear in mind, like you don't always have to use a Pro model. Like even the Flash model can build decent stuff. This was built with Pro, as you can see right here, kind of can build like some cool 3D games just as a demo. So you can get a feel for, you know, how it codes, how it thinks, the logic behind it, etc. And then if you're wondering, okay, how does this compare against Opus 5? So if we look at deep, we've got MiMo 71.9 versus Opus 74. At Terminal Bench, MiMo is actually the best in the category, 89.9. On Automation Bench, MiMo 53.1. And then Agents Last Exam as well. These are from OS World. It's pretty cheap as well. So 87 cents per million words out. Flash is 28. So pretty cheap to use. And then you've got Pro Ultra Speed as well, but that's pretty expensive. But up to 20x faster, which is insane. You can see the details on that right here. Again, 20x faster output speed, which is crazy. You can't get that on OpenCode on the free plan. But still, pretty amazing that you can speed up that much. So the cool thing is about this, like normally people, you know, they're paying for the API. But with OpenCode, you can try it out for free. Just test it, see how you think, etc. It's a mixture of experts model. And you can also plug it into Hermes Agent as well. So for example, inside Hermes, we have MiMo Pro and MiMo Flash plugged in as a brain inside Hermes Agent as well. So the cool thing about that is if I just need to switch to a free model, well, I can use Hermes Agent as a profile with Flash or Pro plugged in right there. And I can use those models for free with OpenCode as a brain, right? So we can have separate profiles for different agents. The same, for example, like Hermes, they just released Claude as a plugin inside Hermes Agent, which you couldn't do before. And we can just select Claude Opus now as a dropdown inside our Agent OS system. So the cool thing about this is, well, like you don't need to pay frontier prices to get a frontier model, which is a big difference. And it's pretty easy and simple to use. Now, you might wonder, OK, how does it perform on everything else? So if we go over to Goldie Bench over here, you can see if we scroll down, we'll have that on the leaderboards later as we test it out more. But yeah, pretty cool model. By the way, if you want a full Agent OS system and all our best trainings on AI automation, how to automate your business, get all your agents plugged in like you can see right here, even use the MIMO code system. And we've got OpenCode inside there. So if you want to use OpenCode inside the Agent OS too, we've got it down here so that you can use free models with the system. Then you can check out the AI Profit Boardroom. Link in the comment description or go to the AI Profit Boardroom. So you get the full Agent OS zip file, the Pro or Flash systems as well. You've got every CLI you already pay for inside there. You can use it with free local models. We have four coaching calls a week as well. So you can basically ask questions, get help and support and share screen. And also inside the community, you can ask questions. And I personally answer them daily as well. So you get everything you need to win with AI automation inside there. By the way, this does not win on Program Bench. So Opus is scoring much higher on Program Bench. Terminal Bench 4.0 as well. It's not really near Opus. Opus is scoring 49 versus MIMO is 34.9. So there's a lot of systems where it's not as good. But again, it's just like a free alternative where you can get decent frontier level performance. Similar to, for example, like Quen 3.8, for example. So if you've got a job and it's not like something super important, you could switch it over to Flash or to Pro as a backup instead of using Claude. And if you look at the progression as well, this is pretty interesting. They've had like five releases in the last 17 months, right? They also showcased the self-reinforcement learning system as well, which is pretty interesting. But, you know, MIMO 7B came out last year. December, V2 Flash. March this year, V2 Pro. Then V2.5 Index 26 came out April 2026. And now V2.6 has just dropped today in September. So that's pretty interesting. Also, some people want to say, well, you know, open models, not as good. But I think we've seen all of that change this year, right? Everything has changed when it comes to open source models. It's just a matter of time, I think, before Chinese models like this actually overtake the frontier models, right? Like you can see them getting better and better. You can see them getting closer and closer. And the biggest difference as well, of course, is like you can use them for free too, which makes a huge difference to people out there in terms of switching. You also might think this is technical to use, but it's literally just like you can open up open code and get access to it. So thanks so much for watching. If you want to get the full system with open code, with Xiaomi MyMovie 2.6 Pro built in, all my best systems, automations, we've got like a video agent over here, an SEO agent, we've got loop systems, everything you need to win with AI automation. You can get that inside the AI Profit Volume. Link in the comment description or go to the AI Profit Volume.com. Inside the community, you can ask questions, get help and support in real time. Inside the classroom, you can get access to all of our best trainings over here. In the calendar, you can jump a weekly coaching course. And then in the map, you can meet people in your local area. So feel free to get that. Link in the comment description or go to the AI Profit Volume.com. Thanks for watching.