StepFun is a general-purpose AI assistant platform established by Shanghai Step Star Intelligent Technology Co., Ltd. in 2023, integrating powerful multi-modal basic models. It not only supports text question answering, knowledge search, language learning, creative writing and code generation, but also has image and video generation capabilities (including text → video, image editing, photo question answering and automatic subtitles). In addition, there are modules such as Diligence Check and Agent Studio, covering daily office, learning research, content production, creation sharing and other scenarios. Accessible locally, mobile, Web, and API, and open source code base (such as Step‑Video‑ T2V, Step‑Audio, Step1X‑Edit, Step1X‑3D, etc.)
Preview
Introduction
functional characteristics
Support text processing capabilities such as text question answering, knowledge search, creative writing, and code generation
Multimodal capabilities for image understanding and editing, text-to-video generation, and 3D content generation
Provide end-to-end speech models for speech interaction, dialect recognition, and emotional expression
Integration of Diligence Check information verification and Agent Studio agent development module
application scenarios
Creative aids: generating copy, scripts, music and visual content
Office efficiency: Excel data processing, mass mail, document analysis and summary
Learning research: in-depth literature search, knowledge base construction, academic report generation
Intelligent terminal: GUI intelligent operation on mobile phone, automobile and other devices
suitable for the croud
Content creators, students, professionals and developers
Educators and researchers who need to efficiently handle multimodal content
type of pricing
Basic features are free to use, some advanced features may have restrictions on use
supported langauges
Support Chinese, English, Japanese, Korean, Spanish, Russian and other 19 languages
Support voice interaction in multiple dialects such as Cantonese and Sichuan
technical highlights
Based on self-developed Step series multi-modal large model, including Step-2 trillion parameter MoE language model and Step-1.5V multi-modal understanding model
Step-Audio 2 mini speech model outperforms GPT-4o-audio on multiple benchmarks
Step 3.5 Flash Agent model with innovative sparse MoE architecture, considering reasoning speed and intelligence level
usage mode
Provide web version (yuewen.cn), mobile App (Yuewen) and desktop client (Xiaoyue)
API calls are supported, and the code base has been open sourced on platforms such as GitHub.
update history
February 2026 Step 3.5 Flash Agent model released for real-time Agent workflow scenarios
In January 2026, upgrade AI desktop partner "Xiaoyue," add global memory and synchronous application functions, and launch Windows version
