Phi-4 is a 14-billion parameter language model developed by Microsoft Research. It is designed to be a small language model (SLM) that excels at complex reasoning tasks, particularly in mathematics, while also being proficient in conventional language processing. The model is trained on a high-quality blend of synthetic and organic data, with a focus on educational content and reasoning-heavy tasks.