Try MiMo-V2-Pro and MiMo-V2-Omni for Free in Kilo Code
Xiaomi just dropped the expanded MiMo-V2 family, and we're making them free for a limited time
The model landscape just got a massive new challenger. Xiaomi has officially dropped the expanded MiMo-V2 family, and this set of powerhouse models is a sign of what’s to come as global AI development reaches record speeds in the era of the always-on agent.
We’ve integrated them directly into Kilo Code, and to help you find your new daily driver, MiMo-V2-Pro and MiMo-V2-Omni are completely FREE to use in Kilo for a full week. Here is our candid take on the new models, how they stack up against the competition for agentic work, and why you should spend the next week stress-testing them.
Do the models feel slightly familiar? That’s because Xiaomi launched the models in stealth on OpenRouter a week ago.
Shhhh…If you tried the stealth models Hunter Alpha and Healer Alpha this past week, now you know: Hunter Alpha is MiMo-V2-Pro and Healer Alpha is MiMo-V2-Omni ✨
They quickly became top Kilo and OpenClaw performers in stealth mode. Now that they’re out of stealth, it’s time to put them to the test in your daily workflows.
The New Roster: Pro, Omni, and TTS
MiMo models are familiar to our customers. Xiaomi: MiMo-V2-Flash has been immensely popular in Kilo for a variety of use cases, a tiny powerhouse—73.4% accuracy on SWB—with remarkable cost efficiency (Just $0.10 per input token and $0.01 for cached).
With the new Pro and Omni releases, Xiaomi is proving they can still help you do more with less. They aren’t just releasing a single model here, they’re dropping an entire ecosystem.
MiMo-V2-Pro: This is the heavyweight reasoning model and an ideal fit for any type of agentic work. Built with a staggering 1 Trillion parameters, Pro is a frontier intelligence model designed explicitly for agentic use. It features a massive 1M token context window, making it excel at long-horizon planning, complex reasoning, and sustained multi-step task execution. It brings the exact kind of instruction-following precision and reliability that frameworks like OpenClaw demand. Pro is exceptionally well-suited for executing complex, multi-step tasks across different modalities in KiloClaw.
MiMo-V2-Omni: This model serves as the versatile multi-modal powerhouse. True to its name, Omni is a frontier omni-modal model equipped with vision, hearing, reasoning, and action capabilities. It features a healthy 262K context window and brings agentic intelligence into the real world by natively perceiving visual and audio inputs.
The TTS Model: Xiaomi also released a lifelike Audio/Text-to-Speech model alongside this update. While the voice capabilities are incredibly cool, only Pro and Omni are free in Kilo for this week. We are keeping our focus strictly on helping you ship code faster.
How Does MiMo-V2 Stack Up?
Let’s be honest: the original MiMo models were blazingly fast but sometimes struggled to hold context when buried deep in a monolithic codebase, or with several passes in agentic work.
V2 completely flips the script, and Pro and Omni are a major leap forward even from Flash. By leveraging Multi-Token Prediction (MTP)—which drastically increases output speed during inference—MiMo-V2 is jumping straight into the big leagues.
With MTP, instead of predicting one token at a time MiMo uses a native draft model to generate multiple tokens in parallel, achieving a 2.0–2.6x speedup and up to 3.6x accepted token length.
On PinchBench, early testing of MiMo-V2 Flash and Pro models found them competitive not just with the latest open-weight powerhouses, NVIDIA’s Nemotron 3 Super and GLM-5-Turbo, but also SOTA models from frontier labs like Anthropic and OpenAI. And the post-launch testing is even more impressive!
Enter the Hackathon
To celebrate the launch of MiMo-V2, we’re running a hackathon where we’re challenging developers to build the “worst captcha ever” using Kilo with Mimo-V2 Pro or Omni.
First place prize is $500 of Kilo Credits and $250 to Amazon.
You know those troublesome widgets that ask you to prove that you’re human? We want you to make one that is so absurd, a robot could never conceive of it.
This is following our “Worst Website” hackathon, where over 500 developers spent a week building terrible websites with Kilo’s App Builder. Turns out there are only a few ways to do something right, but a LOT of ways to do something wrong.
Join the challenge here to win $1250 of total prizes:
Join the Conversation
Building with cutting-edge models is a team sport. Xiaomi has cultivated a highly active developer community around the MiMo ecosystem. If you want to dive into their official documentation, read up on the architecture, or join their specific developer hubs.
But we want to hear from you locally, too. If you want to talk about how MiMo-V2 specifically performs inside your Kilo IDE, the Kilo Discord is a must-join. It’s our community water cooler for comparing models, sharing KiloClaw prompt strategies, and arguing about whether Pro or Omni is the true champion for backend refactoring.
MiMo-V2-Pro and MiMo-V2-Omni are live in your Kilo model picker right now.
The free week starts today. Go break things, let MiMo fix them, and see if Xiaomi’s latest can dethrone your current default model.