Step Fun (StepFun)

Step Fun is a Shanghai-based artificial intelligence lab building frontier foundation models. Its flagship outputs are large language models and multimodal systems, headlined by Step-2—a trillion-parameter model—and Step-1V, a vision-language model that processes and reasons across text and images. The lab primarily serves enterprise customers and developers seeking API access to state-of-the-art Chinese-language and multimodal capabilities, positioning itself alongside DeepSeek, Moonshot AI, Zhipu AI, and MiniMax in China’s intensely competitive generative AI landscape.

Despite being founded in 2023, Step Fun has moved aggressively on model scale, claiming performance on academic benchmarks that places it in the top tier of domestic labs. Its models are expected to power applications in sectors where Shanghai has industrial density—finance, automotive, and smart cities—though the company is still in the early phases of converting technical wins into repeatable revenue streams. The team’s background in large-scale systems research and natural language processing underpins its engineering culture, which prioritizes dense parameter counts and multimodal integration over narrow vertical wrappers.

Step Fun was founded in Shanghai in 2023 under the name Jieyue (阶跃星辰) by Jiang Daxin, a former Microsoft Research Asia vice president with a long record in natural language processing and search. The founding team brought experience from Microsoft and other top-tier technology firms, giving the young lab unusual technical credibility from day one. After operating under the Jieyue name, the company rebranded to Step Fun as it sharpened its public identity around a “step-function” ambition for artificial general intelligence.

The lab’s early trajectory was defined by rapid model releases: it shipped the Step-1 language model family, followed by the multimodal Step-1V and then the massive Step-2, which claims a trillion parameters and has been widely circulated on public leaderboards. In late 2024, Step Fun closed a $200 million Series A at a roughly $2 billion valuation, a significant vote of confidence for a company barely a year old. That financing marked its transition from a research project to a capitalized contender, though it remains one of the youngest entrants in a field where rivals have been shipping products for several years.

Step Fun enters a Chinese foundation-model arena that already includes well-funded incumbents. DeepSeek has captured global headlines for training efficiency and cost-performance, while Moonshot AI’s Kimi chatbot dominates consumer mindshare and enterprise partnerships. Zhipu AI, MiniMax, and Baichuan each have longer operating histories, larger teams, and established distribution. Against this field, Step Fun’s clearest edge is raw scale: a trillion-parameter model and a multimodal system that benchmark competitively despite the company’s youth.

Where Step Fun still lags is in proven product-market fit and brand awareness. DeepSeek and Moonshot have demonstrable user bases and revenue pipelines; Step Fun’s story is currently told more through leaderboard scores than customer case studies. Its path forward likely depends on either securing anchor enterprise contracts in Shanghai’s finance and automotive sectors or building a breakout consumer application that can distribute its models. Without one of these, it risks being viewed as a technically impressive but commercially unproven research outfit.