description = "StepFun's Step models target fast multimodal agents, pairing visual understanding, search, coding, and tool orchestration."