MiMo-V2-Pro

Xiaomi built MiMo-V2-Pro in under a year, creating a trillion-parameter large language model that arrived under the codename "Hunter Alpha" and immediately drew attention for performance rivaling the industry's most advanced closed systems. The model utilizes a 7:1 hybrid ratio—scaled up from the 5:1 ratio in its Flash sibling—to manage its expansive one-million-token context window, enabling it to process and reason over extremely long documents, codebases, or multi-session conversations. This architectural design positions the model for complex reasoning tasks and extended-context applications where depth and continuity matter more than brief interactions.

Before its March 2026 reveal, the anonymous Hunter Alpha model was consuming approximately 500 billion tokens per week on OpenRouter, with developers worldwide speculating it might be DeepSeek V4. Upon Xiaomi's official announcement, the model topped AI rankings in blind tests, cementing its standing alongside frontier models from established players. The company is integrating MiMo-V2-Pro across its broader product ecosystem, signaling a push toward commercial deployment at scale. The model's trajectory—from mysterious challenger to confirmed flagship—reflects Xiaomi's ambition to become a serious contender in foundation model development beyond its hardware origins.