Move over, Claude and ChatGPT: This Android brand's open-source AI model builds 3D games on demand

Smartphones and tablets are likely to come to mind when you first think of Xiaomi. If you’re savvy enough, you might as well know about the company’s diverse portfolio of home appliances and its recent ventures into the EV space. But it might surprise you to learn that Xiaomi is also developing its own frontier AI models to take on the likes of DeepSeek, Claude, ChatGPT, Gemini, and the others.

Xiaomi has just released the latest version of its omnimodal AI models — MiMo-V2.6 Pro and MiMo-V2.6-Flash. These models can be used for textual, visual, and video-based applications, as well as for coding. While the MiMo-V2.6 Pro is Xiaomi’s most capable model, Flash is designed for lower costs and swift responses.

As per Xiaomi, MiMo-V2.6 Pro is the highest-ranking open-source model on the Artificial Analysis Intelligence Index benchmark. It is placed ahead of Qwen3.8 Max and is only slightly behind Claude Opus 5, GPT-5.6 Sol, and Meta Muse Spark, all of which were tested for max effort.

In addition to its open-weights development, Xiaomi is also open-sourcing 7,000+ reinforcement learning (RL) environments as well as the end-to-end RL framework.

Xiaomi adds that the model is capable of 3D spatial reasoning and is flaunting its capabilities through something called “Vibe World.” Using just an image, video, or text prompt, MiMo-V2.6 can build interactive 3D environments such as games, generate models for Blender, or use a multi-camera feed to control a Franka Panda robotic arm. Beyond visual assets, the model can also generate music and compose demo-quality pieces.

Xiaomi is also open-sourcing both models and releasing them via HuggingFace, OpenRouter, and APIs with the same costs as V2.5 Pro. Xiaomi is also widely releasing its MiMo desktop apps for Windows and macOS.