Xiaomi Launches Three MiMo-V2 AI Models — Trillion-Parameter Flagship, Multimodal, and TTS
Xiaomi made it official. After weeks of speculation surrounding an anonymous model called "Hunter Alpha" that appeared on OpenRouter and rapidly climbed its usage charts, the consumer electronics giant formally unveiled the MiMo-V2 family — three models that collectively announce Xiaomi's emergence as a serious frontier AI lab. The flagship, MiMo-V2-Pro, runs 1 trillion total parameters with 42 billion active via a mixture-of-experts architecture and a 1-million-token context window. Built by former DeepSeek researcher Luo Fuli, the model ranks 7th globally on the Artificial Analysis Intelligence Index and 3rd on PinchBench/ClawEval — trailing only Claude Opus 4.6. It's priced at $1/$3 per million input/output tokens, compared to $3/$15 for Claude Sonnet 4.6.
The announcement covered more than the Pro model. MiMo-V2-Omni is a multimodal agent model built to handle images, audio, and complex tool use for agentic workflows. MiMo-TTS brings expressive speech synthesis to the family, targeting Xiaomi's growing hardware ecosystem of phones, smart home devices, and humanoid robots. The company has also committed $8.7 billion in AI investment over three years, per Reuters — a signal that this isn't a research demo, but an infrastructure play.
The deeper story here is what this means for the competitive landscape. When a consumer hardware giant best known for affordable smartphones can field a trillion-parameter model that competes neck-and-neck with Anthropic's flagship at a fraction of the cost, it changes the calculus for every enterprise paying top-tier API pricing. Xiaomi has indicated it plans to open-source a variant of MiMo-V2-Pro once the model is stable — which would put another powerful open-weight option into the hands of organizations looking to reduce their dependence on any single model vendor.