Step Router v1

Step Router v1 is described in StepFun's Step Plan documentation as an intelligent routing model that automatically switches between the deepseek-v4-pro and step-3.7-flash engines for each request. Rather than acting as a standalone generative model, it functions as a traffic director that picks the most appropriate reasoning backend based on the prompt. The Step Plan integration page lists it alongside other supported reasoning models accessed through the unified /step-plan/v1/ endpoint on api.stepfun.com, with subscription and API key prerequisites handled at the platform level.

The practical value of this router lies in letting teams mix a heavyweight general reasoning engine with StepFun's flagship step-3.7-flash, which the documentation characterizes as a multimodal reasoning model supporting native image and video understanding and three reasoning intensity levels (low, medium, high). By dispatching queries to deepseek-v4-pro for harder tasks and to step-3.7-flash for faster or multimodal work, applications can balance cost, latency, and capability without manually picking a model per call. This makes Step Router v1 a good fit for agent systems and mixed-modality pipelines that need automatic selection rather than a fixed model choice.