Updated

Step 5 Preview

StepFun released Step 5 Preview, its flagship agentic model with 600B total parameters, 27B active parameters, and a 1M-token context window. It is available through the StepFun API, and open weights are now hosted in StepFun’s official Hugging Face repository.

By No Code MBA · Facts checked

Provider
StepFun
Announcement date
Status at last check
Generally available
Model family
Step 5
Capabilities
Text, Image, Coding

What changed

Step 5 Preview is StepFun's flagship agentic model, a sparse mixture-of-experts with 600B total and 27B active parameters and a context of up to 1M tokens shared between input and output. The model card lists text and image input with text output and claims frontier-level results in software engineering and professional knowledge work, especially finance, on GDPval-AA v2.1 and DeepSWE v1.1, though the scores appear only in a chart.

Access and availability

Open weights are on Hugging Face (stepfun-ai/Step-5-Preview-BF16, plus an FP8 checkpoint) under the StepFun Model License 1.0, an Apache-2.0-style license with attribution and commercial-license conditions. It is also offered through the StepFun API Platform, and self-hosting is documented with vLLM, SGLang or Transformers.

Official access documentation

What you can build

Evaluate it for long-horizon coding and agentic workflows through the StepFun API. The 1M context window makes it suited for tasks that require sustained context across many turns.

Limitations

Still labeled a preview. Self-hosting needs at least 4 B300 or 8 H200 GPUs for FP8, StepFun says 1M-token requests are not guaranteed on listed configurations, and the two-node BF16 Transformers path is not validated end to end.

Sources and release history