Chinese social media, food-delivery and online travel platforms are increasingly developing their own artificial intelligence models rather than relying on outside providers, betting that combining their proprietary data with tailored AI will create synergies. While the upfront costs are high, some analysts say the approach can pay off over the long term.
Xiaohongshu, a lifestyle-information platform often called China's answer to Instagram, this month released an open-weight model called dots.llm3 Note Preview, with 280 billion parameters, through its in-house AI research unit dots.studio, according to the Hong Kong-based South China Morning Post on the 25th.
"Performance Not Behind OpenAI or DeepSeek"
The company claimed that on certain tasks, the model performs on par with models from U.S. firms OpenAI and Anthropic, as well as Chinese rivals DeepSeek and Zhipu AI. Xiaohongshu's move into large-scale in-house AI development, after growing as a platform for sharing fashion, travel and shopping information, reflects a broader shift underway in China's internet industry, the SCMP reported.
Meituan, China's largest food-delivery and lifestyle-services platform, is also continuing to invest in its AI model family, LongCat. The company is working to apply its own AI across various operations, including support for merchants on its platform and services for users.
Trip.com Group, China's largest online travel platform, has developed its own AI model, Wendao, which it uses for tasks such as trip planning and recommending destinations and products.
Chinese video platform Bilibili uses its Index series for video production and voice generation, while major game developer miHoYo has built Glossa to strengthen interaction between characters and users.
"High Upfront Costs, but Favorable Over the Long Run"
Non-AI companies are increasing their AI investment because AI has become a core element of running a business, said Su Lian-jie, a senior analyst at market research firm Omdia. Tailored AI models backed by a company's own business data can be used to improve efficiency, forecast demand and find new revenue streams, the analyst explained.
Still, some point out that building the infrastructure for an in-house AI model is far from cheap. There are also concerns that companies could struggle to secure specialized talent or to integrate AI with existing systems. "In the short term, this can strain a company's budget," Su said, "but the long-term impact will be very positive."
