Lei Jun posted on Weibo today (March 19) saying that Xiaomi's newly released trillion-parameter model, MiMo-V2-Pro, has climbed to No. 8 on the global Artificial Analysis leaderboard for large models. He also casually mentioned that Xiaomi is putting more than 16 billion RMB into AI research and investment this year alone.
Later in the day, Xiaomi officially launched three new self-developed models:
MiMo-V2-Pro - the flagship text model, built specifically for heavy-duty Agent tasks. It's really good at reasoning, making plans, and calling tools - basically the workhorse for serious AI agent stuff.
MiMo-V2-Omni - a full multimodal beast that handles text, images, and audio all at once. It connects understanding (seeing/hearing/reading) straight through to actually doing things.
MiMo-V2-TTS - their text-to-speech model, designed to make AI voices sound warm, emotional, and human-like instead of that classic flat robot tone.
Both MiMo-V2-Pro and MiMo-V2-Omni are already open for API access right now.
In short, Xiaomi basically built the whole Agent pipeline themselves - from brain to eyes & ears to actual voice output. With 16 billion RMB on the table, they're clearly not messing around in the AI game anymore.
