Step 5 Preview
Step 5 Preview combines a 600B-parameter mixture of experts with 27B active parameters. It accepts text, images and video for long tasks that use tools and documents.
Preview
Work across documents and images
The API accepts up to 60 images per request. It supports JSON Schema, tool calls and three reasoning levels. Tools and external access come from the application that hosts the model.StepFun step-5-preview guide
Read the conditions for your use
The license requires attribution after a commercial product exceeds 100 million active users or US$20 million revenue in a calendar month. Separate commercial terms apply to hardware features and to commercial use by model-service operators above the stated US$20 million annual group revenue threshold. Internal use and certified distribution channels have defined exceptions.StepFun Model License 1.0
Plan the full tool loop
Practical guidance. A large context can hold more evidence, but it does not make external actions automatic. Supply the search, file and execution tools your task needs. Keep the API's output limit separate from local inference settings when you compare results.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Preview |
| Access | API, Downloadable, Provider product |
| Model identifiers | step-5-previewStep-5-Preview-BF16Step-5-Preview-NVFP4 |
| Context window | 1,000,000 tokens The provider specifies up to 1M tokens; account for input and generated output within the model context. StepFun Step-5-Preview model card |
| Hosted output | The API lists a maximum output of 64K tokens. This is a service limit; the weight card describes a larger model-side output capacity.StepFun step-5-preview guideStepFun Step-5-Preview model card |
| Weight formats | FP8, BF16 and NVFP4 are formats of the same release.StepFun official published model repositories |
| Weight license | StepFun Model License 1.0. Device manufacturers and certain large model-service businesses need a separate commercial agreement.StepFun Model License 1.0 |
Related models
- Step 3.7 Flash
StepFun · Step 3
A visual reasoning model with a 196B language backbone and 1.8B vision encoder for images, video, code and tool calls.
Launch date not confirmedAPI · Downloadable - Step 3.5 Flash
StepFun · Step 3
Step 3.5 Flash is a text reasoning model for code and tool use.
Launch date not confirmedAPI · Downloadable