On August 24, sports technology company KEEP (03650) released its first-half 2026 results. During the earnings call, CEO Wang Ning stated that in the first half of 2026, the company completed the launch of Keepace.ai 1.0, its proprietary vertical large model for sports and health, and deployed it across multiple core app scenarios. In August, Keepace.ai officially passed both national and Beijing municipal large model filings, opening full commercialization channels for both consumer and business segments.
Wang Ning first explained the logic behind developing a proprietary vertical model. The sports and health industry has long been stuck on the difficult triangle of "professionalism, personalization, and low cost." General large models tend to have issues with safety, knowledge sources, and personalized matching in sports scenarios. "In this specific context, if a training prescription is wrong, the cost is not just a bad answer, but user injury." Keepace.ai integrates long-term real exercise data accumulated on the platform, sports science literature, and industry expert knowledge, performing post-training on open-source models. It focuses on three core tasks: course generation, knowledge Q&A, and data interpretation.
Wang Ning stated, "We anchor our long-term value on an AI-driven sports and health ecosystem. It's not about putting an AI shell on our business, but about solving the industry's most fundamental constraints." To validate the model's capabilities, KEEP partnered with Capital University of Physical Education and Sports to release the industry evaluation framework MoveBench. This framework assesses the model's comprehensive execution across the three core tasks from four dimensions: exercise safety, scientific accuracy, personalized adaptation, and actionability of recommendations. Under the MoveBench framework, Keepace.ai performed well in horizontal comparisons against leading industry large models.
Wang Ning noted that the value of KEEP's vertical model is ultimately realized in products, landing in training courses, voice running companions, and full-chain integration. To date, KEEP's exercise course agents cover key categories such as gym equipment, yoga, and running, with over 8,000 AI courses launched in the first half of the year. The voice running companion agent, based on Keepace.ai, enables rich two-way interaction. In the newly iterated Keep App 9.1, AI is embedded throughout the entire exercise journey—before, during, and after workouts.
Even more noteworthy is the business-to-business space. Wang Ning said that sports hardware manufacturers, insurance companies, and medical and health sectors all have demand for AI service integration. KEEP's accumulated three-layer advantage of "sports science resource library + filed vertical model + proven agent capabilities" positions it exceptionally well to become the AI solution provider for the sports and health industry. He cited the application value of sports and health scores in insurance and medical sectors as an example: "If merchants in these industries can call KEEP's interfaces to improve service efficiency and value for their end users, it would be a win-win situation." Following this logic, the commercialization potential of KEEP's proprietary sports and health large model in the B2B space is substantial, with subscription models, project-based models, and usage-based billing all representing viable and promising directions.