StepFun
StepFun develops the Step model families for language, visual tasks and speech. Its catalog includes hosted APIs and published research weights.
Company & leadership
- Chief executive
- Jiang Daxin; confirmed in StepFun's April 2026 cooperation announcement.
- Chairman
- Yin Qi (印奇), appointed in January 2026.
- Main office
- Floor 30, 701 Yunjin Road, Xuhui District, Shanghai, China, as recorded in the May 2026 filing.
- Legal entity
- 上海阶跃星辰智能科技股份有限公司, a Chinese joint-stock company.
- Control
- The May 2026 investment disclosure identifies Yin Qi as controlling shareholder and actual controller. This is a dated ownership statement.
Founder
- Jiang Daxin (姜大昕), identified as founder by investor Qiming Venture Partners.
A Shanghai model company founded in 2023
StepFun's registered company dates to April 2023. Its main office is in Shanghai's Xuhui District. The May 2026 filing records the legal entity and its control structure; it does not establish a live shareholder register.
In a June 2025 interview with investor Qiming Venture Partners, founder Jiang Daxin described a focus on models that understand and generate several types of media. He linked that work to agents for cars, phones and other devices.
Leadership expanded in 2026
Yin Qi became chairman in January 2026. StepFun named Jiang Daxin as CEO, Zhang Xiangyu as chief scientist and Zhu Yibo as CTO in the same announcement. Yin also chaired Qianli Technology, which StepFun identified as a partner for device applications.
Models connect to phones and cars
In April 2026, StepFun and Tencent Cloud announced work on vehicle assistants. StepFun supplied model technology; Tencent Cloud contributed cloud, content and application services. This partnership is distinct from the authorship of either company's model families.
StepFun announced Step Edge in July 2026 for local text, visual, audio, GUI and image tasks. It also introduced an NPU inference engine. The reviewed announcement does not provide general downloadable weights or a public API for each Edge model, so those names are not listed below as public releases.
A joint institute studies agents
In July 2026, StepFun and Shanghai Qi Zhi Institute announced a joint research institute. Its stated subjects include agent networks, their economic principles and AI safety. The collaboration connects model development with research into how agents act and work together.
Public access differs across model lines
StepFun's API catalog focuses on reasoning, visual understanding and audio. Its image-generation APIs ended service on 10 October 2026. Published image and video weights remain separate releases; an API shutdown does not by itself remove those downloads.
Practical guidance. The entries below cover reviewed API releases and published weights. Packaging formats and API interface names stay with the underlying release. Research derivatives name their base models. This guide does not claim to list every old checkpoint.