The Mystery Is Solved

On March 11, two unnamed models — Hunter Alpha and Healer Alpha — appeared silently on OpenRouter with extraordinary specs and no lab attribution. Weeks of speculation followed, with many pointing to DeepSeek as the origin. On March 18, Xiaomi ended the mystery: the models were its unreleased MiMo-V2 series, now officially launched with global API access.

Three Models, Serious Benchmarks

The MiMo-V2 lineup consists of three components:

MiMo-V2-Pro (Hunter Alpha) is a text reasoning model with 1 trillion total parameters, 42 billion activated at inference, and a 1-million-token context window. On the Claw-Eval benchmark, it scored 75.7 — placing third globally, directly behind Claude Opus 4.6. On the Artificial Analysis Intelligence Index, it ranked eighth globally and second in China, edging out Grok 4.20 and Gemini 3 Flash.

MiMo-V2-Omni (Healer Alpha) is a multimodal model natively processing text, image, video, and audio, with a 262K token context window. It topped the PinchBench leaderboard and outperformed Gemini 3 Pro and Claude Opus 4.6 on audio and speech reasoning benchmarks.

MiMo-V2-TTS rounds out the suite with speech synthesis trained on hundreds of millions of hours of audio, enabling precise emotional control and multilingual dialectal output.

Global API, Competitive Pricing

Xiaomi has made the API immediately available worldwide at platform.xiaomimimo.com, priced at $1.00 per million input tokens and $3.00 per million output tokens for MiMo-V2-Pro — a fraction of comparable frontier model costs. Artificial Analysis currently ranks MiMo-V2-Pro as the top cost-effective model in its category.