Phi

Phi is the compact LLM family from Microsoft AI research, built around the thesis that smaller models with very high-quality data can punch above their weight. Heavy use of Synthetic Data is one of its signatures and a key reason it consistently outperforms its parameter class. Later releases added long-context, multimodality, and Reasoning capabilities, and the family became a flagship example of On-Device LLM deployment. Releasing as Open Weight helped it spread quickly across the developer ecosystem.