Step-3.7-Flash

Step-3.7-Flash is StepFun's flagship high-efficiency multimodal reasoning model, built on a sparse Mixture-of-Experts architecture (198B total / ~11B active parameters) that delivers high throughput and low latency for real-time agentic workflows and high-frequency calls. It natively understands images and video — no extra vision model needed in agent frameworks — and offers a 256K context window with three selectable reasoning levels (low/medium/high) to balance speed, cost and reasoning depth. With reliable multi-step tool calling, task decomposition and plan execution, it excels at coding, solution planning and vision-to-workflow tasks, producing complete plans in a single call.