StepFun's Step 5 Preview: An AI Agent Built for 24-Hour Task Execution

StepFun 6 hours ago

Key Info

StepFun introduced Step 5 Preview, a new flagship agentic model that sustains execution across long horizons—tested in runs up to 24 hours—and works across software engineering, web applications, 3D workflows, and programmable hardware.

Highlights

Tracks prior results and uses execution feedback to decide next actions, continuing to iterate without human intervention.
Tested on long-duration tasks including GPU kernel optimization and automated post-training.
Covers a broad range of environments, from software engineering and web apps to 3D workflows and programmable hardware.

Original post

Step 5 Preview works across software environments and sustains execution over long horizons. Its capabilities extend from software engineering and web applications to 3D workflows and programmable hardware. Over longer horizons, Step 5 Preview keeps track of prior results, uses execution feedback to decide what to try next, and continues iterating. We test this behavior in runs lasting up to 24 hours, including tasks involving GPU kernel optimization and automated post-training.

#StepFun#Step 5 Preview#AI agent

View original on X