Step 5 Preview utilizes a Mixture-of-Experts architecture with 600 billion total parameters and 27 billion active parameters. The model targets software engineering and professional knowledge work, specifically in finance, and provides a 1 million token context window. StepFun is currently offering access via API and AI Studio with pricing set at $1.00 per 1 million input tokens and $2.70 per 1 million output tokens.

Early testing of the model as a coding agent places its performance on par with GLM 5.3 and Kimi K3. A user report showed the model identified root causes within 368,000 tokens of incident tickets in 90 seconds and declared a task finished while GLM 5.3 continued until hitting a step limit. StepFun releases the model weights on Oct 15.