Chinese electronics manufacturer Xiaomi launches two multi-modal artificial intelligence models in the series Xiaomi V2.6. The company assures that they can compete with developments from the world’s leading labs, including DeepSeek, OpenAI, Anthropic and Google.
Image source: Xiaomi.com
The new AI model series is represented by the flagship version MiMo-V2.6 Pro and the faster and more efficient version MiMo-V2.6-Flash. Because they are multimodal, they are equally good at generating text and code and processing images and video. The developer guarantees that the flagship product Xiaomi MiMo-V2.6 Pro ranks first in the open model artificial intelligence analysis wisdom index. It’s ahead of Alibaba Qwen3.8-Max and only slightly behind Anthropic Claude Opus 5, OpenAI GPT-5.6 Sol and Meta✴ Use Muse Spark when testing in maximum performance mode.
Xiaomi not only opened up the weights of the new model itself, but also opened up the code of related tools: more than 7,000 reinforcement learning environments, as well as the architecture of a full range of reinforcement learning tasks. The model has spatial thinking capabilities – its capabilities are demonstrated in the Vibe World project. Based on text requests, samples in image or video format, she creates interactive spaces, generates Blender models, and controls a Franka Panda robotic arm using data from multiple cameras. In addition to visual objects, the model can also produce music.
The developers announced the weights of the two models on the platform Face hugging and open routerand provide them via API at the same price as V2.5 Pro. Xiaomi also released client application Use your AI models under Microsoft Windows and Apple macOS.
If you find an error, select it with your mouse and press CTRL+ENTER.
