Phi 3 Vision is a compact multimodal model that punches above its weight class for visual reasoning tasks. It handles image understanding and text together with a notably long 128k context window, making it comfortable with lengthy documents alongside visual inputs. Its open-weight MIT license means it runs locally without restrictions, though it trades raw power for efficiency.