A：In the first half of this year, the business performance of Shangtang Group was remarkable, achieving a revenue of 2.91 billion yuan, an increase of 23.4 percent year-on-year. Among them, the revenue of generative AI was 2.33 billion yuan, up 28.2 percent year-on-year, accounting for nearly 80 percent of the group's total revenue; the revenue of visual AI was about 0.5 billion yuan, up 13.9 percent year-on-year. In addition, overseas business revenue increased by 127 year-on-year, becoming an important driving force for growth. Gross profit reached 1.21 billion yuan, up 32.9 year on year, and gross profit margin rose to 41.4. Recurring revenue (current revenue) was RMB 1.14 billion, up 124.4 per cent year-on-year, accounting for 39.3 per cent of the Group's revenue, reflecting the continued deepening of partnerships with customers and significant improvements in operating efficiency, with adjusted net loss narrowing to RMB 0.39 billion, down 67.3 per cent year-on-year. At the same time, Shang Tang Group has achieved a FS-caliber net profit of 0.62 billion yuan, which is also the first time that Shang Tang has achieved FS core profit since its listing.

A：The technical infrastructure and products of Shangtang Group are forming a more complete business closed loop, which is reflected in three aspects: first, by building a "trinity" core capability system-a set of models, a token factory and a set of intelligent body management and control system, the function is improved and the model's processing capability for text, image, video space and application interface is realized; Secondly, through the joint optimization of model and infrastructure, energy consumption costs are reduced and production efficiency is improved, and finally, the model and token are transformed into complete tasks through the intelligent body control system to support the development of different intelligent products at the front end. This series of initiatives has shifted the AI business model from simply selling models or tokens to measuring value by completing tasks and delivering results.

A：Shang Tang's technical route in model development is consistent and continuous, with particular emphasis on the development of multi-modal technology. In the early stage, image perception and language model combined to achieve the basic multi-modal understanding ability, has now entered the second stage, through the native multi-modal ability and intelligent body effective combination training, to achieve efficient processing of mixed modal information multi-modal model. The 30 nova series model released in April this year has realized unified understanding, reasoning and generation in a single model. The recently released U1.5 model has further penetrated this framework, demonstrated strong expansion capability, and has formed a significant influence in the community. Open source projects have accumulated more than 11000 stars on UP. The unified model architecture integrates visual perception ability, improves visual reasoning and controllable generation effect, and makes important breakthroughs in visual understanding, mixed graphic reasoning and controllable image generation. For example, U1.5 light model performs well on the VBVRP list of authoritative visual reasoning, while maintaining low cost and high efficiency. In addition, Shang Tang also plans to launch a larger parameter Great Wall intelligent body model 306000.8 flash, which is expected to achieve better performance in more complex tasks and multi-intelligent body collaboration.

A：The Shang Tang large device improves yield and reduces costs through model adaptation, reasoning optimization, heterogeneous scheduling, and computing power collaboration, while improving model reasoning, accuracy, reliability, and response speed. In the long run, it will shift from producing tokens to completing tasks, focusing on the quality success rate and cost of unit tasks. One of the key tasks is to support the development and transformation of the Shang Tang model and maintain the leading level of the system support system in the industry. Through technical iteration and innovation, optimize sequence parallel and computational communication coordination strategies to improve training speed, use domestic computing power to speed up sample generation and improve post-training efficiency, self-developed reasoning framework to improve image and video generation efficiency, and achieve rapid large-scale deployment of models.

A：Shang Tang research and development of the full stack adaptation system, including the underlying operator optimization, multi-card parallel tuning and other technologies, so that the mainstream domestic chip MFU to reach 2.5 times the chip original benchmark level, optimize the cost-effective, to protect the domestic chip market profit, shorten the distance from model innovation to commercial services.

A：Shang Tang has cooperated with Hong Kong Science and Technology Park to build the world's largest domestic manufacturing center, with a target of 40000 computing power by 2030. The first overseas domestic computing power cluster in Saudi Arabia has not only expanded the computing power node, but also improved the globalization level of token production and delivery capability.

A：At present, the total force of operation has reached 48000, and the average daily token service call volume has increased by about 22 times to 2.4 trillion times year on year.

A：The Shangtang installation not only serves the new model of Shangtang Ri, but also provides services to four external basic model manufacturers, while optimizing energy efficiency, embracing green energy, and achieving sustainable development.

A：The intelligent body management and control system is responsible for understanding user goals, dismantling tasks, calling knowledge and tools to arrange workflows, unified management of identity permissions, long-term memory, etc., and accessing enterprise-level privatization deployment services through unified intelligent body capabilities to support task completion quality reliability and delivery efficiency in different product industries and user scenarios. Products such as small raccoon office intelligence have achieved remarkable results in different scenarios, with the number of users increasing rapidly. In the field of content production, the video creation capability of cycle platform is continuously improved, and active users and creative content are growing rapidly. In the personal user market, users of the Kapi family product matrix have broken through the 45 million and provided active and professional intelligent services for users' life goals.

A：The visual AI business covers many countries and regions around the world. The income contribution rate of old customers is high, and the product and service system is more complete. It can meet the depth and breadth of customer needs from single-point application to multi-business links. It has remained in China for ten consecutive years. Ranked first in the history of the computer world, and was recognized by the second prize of National Science and Technology Progress.

A：In the first half of 2026, Shang Tang continued to focus on the high-quality development of its core business under the guidance of "a set of models, a token factory, and an intelligent body management and control system. The One Plus X strategy and the AI ecosystem it has built have begun to show positive results in the financial dimension. During the period, we achieved healthy growth in revenue and gross profit, continued to decline in operating costs under refined management, improved scale and efficiency simultaneously, and achieved the first positive change in net profit since listing.

A：In the first half of 2026, Shangtang recorded revenue of 2.91 billion billion yuan, an annual growth rate of 23.4 percent; gross profit reached 1.21 billion billion yuan, an annual growth rate of 32.9 percent, and gross profit margin of 41.4 percent, an increase of 2.9 percentage points over the same period in 2025.

A：Shang Tang continued to maintain strict and efficient operating expenditure control, sales management research and development three expenses have achieved year-on-year decline. Total operating expenses fell 14.8 per cent year-on-year and overall research and development expenses fell 17.1 per cent year-on-year, while working capital flow efficiency improved and the cash cycle was 34 days shorter than a year ago.

A：The impairment provision ratio for trade receivables on Shang Tang's balance sheet decreased significantly, resulting in a negative to positive impairment loss on financial assets. As of the end of the first half of 2026, the company's total cash reserves reached 13.8 billion billion yuan, capital expenditure was mainly used for large-scale computing infrastructure construction, and the total debt level was 25.7 percent lower than a year ago.

A：Shang Tang will focus on three directions around a set of models, a token factory and a set of intelligent body management and control systems: first, to promote the continuous evolution of multi-modal Great Wall intelligent bodies in real tasks; The second is to promote the transformation of the pit-removing factory from large-scale production intelligence to task delivery. The third is to promote multi-agent collaboration into deeper personal and enterprise workflows, system-level generation for coverage understanding, plan execution, validation, and delivery.

A：Recurring revenue (RR) is revenue based on contracts in force during the reporting period and with continuing renewal attributes. In the first half of this year, Shangtang's recurring revenue reached 1.14 billion billion yuan, an increase of 124.4 per cent compared to the same period in 2025, accounting for 39.3 per cent of the group's total revenue.

A：Shang Tang will consistently promote the development of large models, especially native multimodal technology, and believes that this will bring about rapid improvement and large-scale qualitative changes in intelligences. in application scenarios, although seemingly diverse, it is based on the multimodal capability of a transparent factory and agent harness, and there is a synergistic effect in different high-frequency scenarios. The choice of office scenario is due to the fact that when the intelligent body is applied in depth, the customer base is expanded to the general public, the application scenario is more extensive, and the commercialization logic is clear. For coding capabilities, although the small raccoon code products have been online and integrated into the integration of the entrance, but at present is not entirely the focus, because of fierce competition. Code capabilities are continuously integrated into model products to improve the logic and reasoning capabilities of the overall agent.