Tsinghua proposes a capability density evaluation system: AI models' volume halves every 3.5 months, Midea Intelligent has been applied to smartphones, in-car systems, and home appliances

Tsinghua University jointly with Midea Intelligent published an article in "Nature Machine Intelligence", proposing the "capability density" metric, emphasizing that model performance should focus on effective intelligence per parameter rather than scale. The study found that capability density doubles approximately every 3.5 months, allowing the required parameters for the same task to decrease exponentially. High-density requires collaborative design of data, computing power, and algorithms, rather than simple compression. Based on this, Midea Intelligent has introduced a series of models ranging from 0.5B to 2B.