StepFun

StepFun (阶跃星辰, also referred to as Jieyue Xingchen) is a Shanghai-based AI foundation model startup founded in 2023 by Jiang Daxin, a former Microsoft Research Asia chief scientist and 16-year Microsoft veteran who worked on Bing, Cortana, Azure and Microsoft 365 natural-language systems.

The company builds the Step series of foundation models spanning language, multimodal vision, image, video, audio and agent products, including Step-1, Step-2 (a trillion-parameter mixture-of-experts model), the Step-Video-T2V text-to-video model, Step-Audio 2, and the more recent Step 3.7 Flash, a 196-billion-parameter multimodal model built for agentic, real-world tasks with native image/text understanding, web and visual search, and tool use.

StepFun describes its strategic focus as "physical AI" and an "AI plus terminal devices" push into smart vehicles, personal devices and embodied intelligence, and its models have reportedly been adopted by device makers including OPPO, Honor and Geely-linked auto ventures.

In January 2026 Megvii co-founder and former CEO Yin Qi, who also chairs the Geely-backed smart-driving firm Qianli Technology, joined StepFun as chairman alongside a large funding round; the company has since dismantled its offshore "red-chip" corporate structure, a step commonly taken by Chinese companies preparing a Hong Kong listing, and is reported to be planning a Hong Kong IPO. StepFun is grouped by Chinese press among the so-called "AI Six Tigers" cohort of Chinese foundation-model startups, two of which (Zhipu and MiniMax) had already completed Hong Kong listings by January 2026.