At Xiaomi's 2025 Human-Car-Home Ecosystem Partner Conference held on December 17, Luo Fuli, Head of Xiaomi's MiMo Large Model team, made her first public appearance to introduce the open-source MiMo-V2-Flash.

The new model is designed to be the backbone of Xiaomi’s "Human-Car-Home" ecosystem, emphasizing efficiency and specialized Agent collaboration.

Luo Fuli highlighted three core pillars that define the model's technical edge:

Superior Tool Integration: Engineered as a high-efficiency "language" for AI Agents, boasting exceptional code generation and tool-calling capabilities.

Optimized Inference Architecture: Designed for "high-bandwidth" collaboration, allowing multiple Agents to handle complex tasks with minimal latency.

Next-Gen Post-Training: Utilizes a stable, scalable Reinforcement Learning (RL) paradigm to continuously refine performance without massive compute overhead.

Despite its relatively lean architecture—309B total parameters with only 15B activated—the model punches significantly above its weight class.

The debut marks a significant milestone for Xiaomi's AI ambitions. Luo Fuli, a prominent AI prodigy formerly associated with the development of DeepSeek-V2, officially joined Xiaomi in November 2025.