Since ChatGPT, large language models have split into two camps. General models such as Baidu’s ERNIE, Alibaba’s Qwen and iFlytek’s Spark target the base layer, while vertical models train industry-specific systems on top of general ones for finance, health care, education and transport.
Wu Bingkun, chief executive of Zhongshu Xingke, says from the client’s side the thing that matters most is customisation and the vendor’s ability to deliver as engineering, not raw model power. His company, founded in early 2021 by CloudWalk, Xiamen Torch Venture Capital and Minsheng Commerce, positions itself as knowledge intelligence, fine-tuning general models into industry models for digital cities.
Wu argues the big labs will not open their ability to fine-tune general models into industry models, because they are still shoring up base capability and are not yet mature at engineering deployment. That gives vertical players a window, but using a general model as a base and customising it is far from trivial.
Zhongshu’s edge is the three factors that matter in deployment: research, reachable industry scenes, and high-quality industry data. It touches 30 million individual users and hundreds of thousands of enterprise users across six provinces and 16 cities through its i-City platform, and its commercial logic is to enter through government, bank experience in the business segment, and replicate fast on the consumer side.
In education, built on CloudWalk’s Congrong model, it ships a teacher’s digital assistant that auto-generates lesson plans and personalised assessments, now piloted in Xiamen, with a longer road toward a teacher’s digital twin. Wu warns that a breakout product likely creates a winner-takes-all outcome, and vertical players race against time as crowded entrants such as iFlytek and Yuanfudao arrive.
The strategic question mirrors the Western market exactly. The general-model giants pour capital into scale, while smaller firms bet that owning a vertical workflow and its data is the defensible position. Both routes are still running, and neither has won.
Editor’s note: This is an adapted translation of the original LeiPhone report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://www.leiphone.com/category/ai/Zt5mFmrAgQP43K9x.html.