Step 3.7 Flash is a 196B-parameter model designed around the See.Think.Act. approach for real-world agents. It combines visual understanding of interfaces, documents, charts, and natural scenes with code generation and tool-driven action, making it well suited to workflows that move from inspection to execution rather than producing text alone.

Its practical focus is sustained agent work: it can operate terminals, browsers, office tools, and search, while maintaining coherent multi-step tool use. Enhanced web and visual search supports broader source discovery, deeper follow-up, and recognition of long-tail or newly emerging entities. The model is also offered as an NVIDIA NIM, giving organizations a supported deployment path alongside its public-weight distribution.