Capability density

The empirical trend (the 'densing law') that the capability achievable per model parameter roughly doubles over time, so ever-smaller models match older large ones and become runnable on commodity, GPU-less hardware.

grounded in: Trend theme 'Running big LLMs on modest/local hardware' + Notable: 'Gemma 4 26B (local): running a 26B model on a decade-old CPU with no GPU'.

Connected concepts

Scaling laws, Small language model, CPU inference serving, Capability–cost frontier