罗福莉首秀，雷军的AI新战事

Core Insights - Xiaomi has introduced the MiMo-V2-Flash model, which is designed to serve as a foundation for AI agents, emphasizing high cost-effectiveness and speed [1][5][21] - The model boasts 309 billion parameters and a generation speed of 150 tokens per second, positioning it among the top open-source models globally [1][11] - Xiaomi's AI strategy focuses on lightweight models and local deployment, aiming to enhance its smart terminal capabilities and intelligent driving features [5][6][25] Model Performance - MiMo-V2-Flash has demonstrated superior code capabilities, reportedly surpassing several closed-source models, including GPT-5, in specific benchmarks [11] - The model's inference cost is only 2.5% of that of Claude Sonnet 4.5, while its generation speed is double, making it competitively priced in the market [12][21] - The architecture employs a hybrid attention mechanism, combining sliding window attention and global attention to optimize performance [14] Strategic Importance of AI - The launch of the model at the "Human-Vehicle-Home Ecosystem Conference" underscores the strategic significance of AI for Xiaomi [4][18] - Xiaomi plans to invest over 300 billion yuan in R&D by 2025, with a quarter of that amount allocated directly to AI development [21][25] - The company aims to leverage its extensive ecosystem of over 1 billion connected devices to enhance its AI capabilities and user experience [25][26] Future Directions - Xiaomi's future optimization efforts will focus on two main areas: enhancing smart terminal capabilities and developing physical models for intelligent driving [15][25] - The company is building its own AI infrastructure and GPU clusters to support its ambitious AI initiatives [21][25] - Xiaomi's dual-core approach, with key personnel focusing on foundational models and intelligent driving, indicates a strategic alignment towards comprehensive AI integration [21][25]