Step 3.7 Flash

Step 3.7 Flash is StepFun’s open-weight sparse mixture-of-experts vision-language model for agentic workflows, coding, search, tool use, and image understanding.

It fits tool-using agents, repository-scale coding, visual UI and document analysis, research workflows, and other high-throughput tasks that combine perception and reasoning.

Native vision, long-context processing, three reasoning levels, a focus on stable tool use, and open deployment options make it flexible for production agent systems.

It is a very large model to self-host, so local deployment can require substantial memory and compute. Its outputs can still be wrong; validate important results and compare quality, latency, and cost on your own workload.

Choose Step 3.7 Flash as the model in your TokenHub API request, then use the endpoint and credentials shown in your TokenHub account. Follow TokenHub’s current guidance for supported text and image payloads.

TokenHub pricing, regions, and quotas can change, so check the live model listing and your account before deployment. The model also has open weights and upstream hosted options, while TokenHub availability is governed by TokenHub.

Choose it when you need a fast multimodal agent model with tool use and open deployment choices. For simple text-only tasks, compare a smaller model; for maximum accuracy, benchmark stronger alternatives on your own data.