Step 5 Preview
Step 5 Preview is StepFun’s flagship open-weight multimodal model for software engineering, long-horizon agentic tasks, and professional knowledge work, with 600B total parameters, 27B active per token, and up to a one-million-token context window.
Model
Step 5 Preview is StepFun’s flagship model for agentic work. Its sparse Mixture-of-Experts architecture has 600B total parameters and activates 27B per token, with support for up to one million tokens of context and native text and vision inputs.
Built for software engineering and professional knowledge work, Step 5 Preview can navigate unfamiliar codebases, coordinate changes across files, and iterate through tool use, execution, and verification. It can turn screenshots, mockups, or product requirements into working web pages, and analyze annual reports and industry material to produce financial reports, editable spreadsheets, and slide decks.
The model uses a 92-layer architecture and Sparse GQA to support long-context reasoning. Its reinforcement learning incorporates context compression into policy training for long-horizon tasks. StepFun also provides Step Code, an open-source agent harness built around Step 5 Preview.
Step 5 Preview is released under the StepFun Model License 1.0.
Dedicated Endpoint
DocsA Dedicated Endpoint is your own deployment of Step 5 Preview, on an inference stack that Modal tunes and autoscales. Bring fine-tuned weights if you have them. Compute is billed by the second, and only while it runs.
modal endpoint create --model stepfun-ai/Step-5-Preview