MODELS

StepFun: Step 3.7 Flash

by StepFun

modellead-sourceopenrouter-modelssource:github.commultimodalvision-language-modelmixture-of-expertslong-contextstepfun

Overview

High-efficiency multimodal sparse MoE vision-language model from StepFun, available through StepFun API and chat interfaces.

Details

Step 3.7 Flash is StepFun’s high-efficiency multimodal Mixture-of-Experts model. Source descriptions identify it as a sparse MoE vision-language model with a roughly 196B/198B-parameter scale, native image and video understanding through a vision encoder, around 11B active parameters, a 256k context window, selectable reasoning levels, and local deployment instructions in the official GitHub repository. StepFun’s official page says it is available through StepFun’s API platform and web/app chat interfaces.

When to Use

Use when you need a StepFun multimodal model for text, image, or video understanding workflows. Use when long-context tasks may benefit from the model’s stated 256k context window. Use when you want selectable reasoning levels or a high-efficiency sparse MoE vision-language model to evaluate.

Getting Started

Read the official StepFun model page: https://static.stepfun.com/blog/step-3.7-flash/
Review the official GitHub repository for model details and local deployment instructions: https://github.com/stepfun-ai/Step-3.7-Flash
Access the model through StepFun’s API platform or web/app chat interfaces, as described by StepFun.
Optionally compare third-party availability and metadata on OpenRouter: https://openrouter.ai/stepfun/step-3.7-flash

Key Features

•Sparse multimodal Mixture-of-Experts vision-language architecture
•Native image and video understanding via a vision encoder
•Approximately 196B/198B total parameter scale with roughly 11B active parameters
•Stated 256k context window
•Selectable reasoning levels
•Official GitHub repository includes local deployment instructions

Capabilities

•text-generation
•vision-language-understanding
•image-understanding
•video-understanding
•long-context
•reasoning-level-selection
•local-deployment

Last updated May 29, 2026