Step 3.5 Flash
Step 3.5 Flash is a text reasoning model for code and tool use. The published architecture has a 196B language backbone, approximately 11B active parameters and a separate prediction head.
Separate serving variants from training stages
The API guide groups the original and 2603 variants together. The published Base and Base-Midtrain checkpoints are separate training stages. FP8 and GGUF files are deployment formats.StepFun step-3.5-flash guideStepFun Step-3.5-Flash model card
Use text input deliberately
Practical guidance. This model fits tasks whose evidence is text, code or tool results. For a workflow that must inspect an image directly, compare Step 3.7 Flash. A conversion of a screenshot to text can lose layout information that matters to a task.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API, Downloadable |
| Model identifiers | step-3.5-flashstep-3.5-flash-2603Step-3.5-Flash-FP8Step-3.5-Flash-GGUF-Q8_0Step-3.5-Flash-GGUF-Q4_K_S |
| Context window | 256,000 tokens Provider-listed 256K context window. StepFun step-3.5-flash guide |
| License | Apache 2.0.StepFun Step-3.5-Flash model card |
| Attention design | Three sliding-window attention layers for each full-attention layer.StepFun Step-3.5-Flash model card |
| API variant | step-3.5-flash-2603 adjusts token use and agent integration. It supports low and high reasoning effort.StepFun step-3.5-flash guide |
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 Base
StepFun · Step 3
A pretrained sparse language-model checkpoint from the Step 3.5 Flash family.
Launch date not confirmedResearch releaseDownloadable - Step 3.5 Flash Base Midtrain
StepFun · Step 3
A midtraining-stage checkpoint of the Step 3.5 Flash sparse language model.
Launch date not confirmedResearch releaseDownloadable