After the token revenue surged by 760%, the market is re-evaluating UNISOUND (09678).

UNISOUND (09678) is entering a moment worth reevaluating. Lets start with some numbers. In the first half of 2026, UNISOUND achieved revenue of 562 million yuan, a year-on-year increase of 38.7%; revenue from the large model Token business approached 30 million yuan, with a year-on-year increase of 760%; of this, the revenue for the second quarter alone exceeded 25 million yuan, with a quarter-on-quarter increase of over 500%. Even more noteworthy is that the rapid growth of this round of Token revenue is primarily driven by income in US dollars. This changes the significance of the 30 million yuan figure. If the growth mainly comes from traditional domestic projects, it reflects more about order expansion; however, the increase in US dollar Token revenue corresponds to an entirely different business modelusers directly access the model API and pay based on Token usage, thus the model's capability itself begins to become a commodity. At the same time, the gross margin of the Token business has exceeded 60%, and the proportion of recurring revenue for the company has also surpassed 60%. On another front, UNISOUND is still ramping up R&D, with R&D investment in the first half of the year reaching 284 million yuan, a year-on-year increase of 69%, yet losses during the same period continued to narrow. When these curves are put together, a previously less apparent change begins to surface: UNISOUND is gradually revealing the revenue characteristics of a large model company, moving away from being perceived by the market purely as an AI project company. This may be the most significant aspect of this interim report. In the past few years, the large model industry has always been answering one question: Whose model is stronger? By 2026, the capital market began to pose the next layer of questions: Whose model is actually being used? Who can enter real workflow? Who can turn model usage into sustainable revenue growth? Looking at UNISOUND from this perspective, a revenue growth of 38.7% is not the most interesting number in this financial report. What truly deserves attention is the Token. Because it is providing a new yardstick for the market to reevaluate this company. A new yardstick appears; the curve of Token driven by US dollar income has steepened. In the past six months, UNISOUND's most tangible change comes from the Token. In the first half of the year, revenue from the large model Token business approached 30 million yuan, a year-on-year increase of 760%; among this, the revenue for the second quarter alone exceeded 25 million yuan, with a quarter-on-quarter increase of over 500%. This means that the vast majority of Token revenue for the first half of the year was generated in the most recent quarter, and the growth curve is clearly becoming steeper. Moreover, this round of growth is primarily driven by US dollar income. For a large model company, this fact is crucial. Because Tokens and traditional AI projects operate under entirely different business logic. Traditional software and AI projects rely more on procurement, customization, and delivery, with each contract corresponding to a piece of revenue; Token revenue occurs only when the model is truly accessed. Only when there are calls can there be revenue; the more calls, the higher the revenue. Thus, what lies behind Token revenue reflects not merely that the model has sold, but that the model is being continuously used. Especially entering the Agent era, this logic will be further amplified. In the past, interactions between humans and AI were primarily question-and-answer, with one question corresponding to a handful of model calls. But as agents start to perform complex tasks, a workflow might be broken down into dozens or even hundreds of steps, each step potentially calling models, tools, or external systems. The unit of model consumption is shifting from one conversation to completing one task. UNISOUND U2 can autonomously decompose and advance complex workflows of over 100 steps, precisely adapting to this change. Therefore, what is truly worth noting is not the 30 million yuan itself, but the nature of the revenue behind this curve. At present, the gross margin of the Token business has exceeded 60%. This indicates that UNISOUND is not merely relying on low prices to gain usage but is forming a new revenue curve characterized by high growth, high gross margin, and recurring features. Project revenue resembles one-time sales. Token revenue implies continuous usage. When models begin to earn money based on use, the business logic of an AI company fundamentally starts to change. A transformation from projects to assets. But the Token is merely the result. What truly underpins this curve is another change occurring within UNISOUND. In the past, the common perception of UNISOUND was as an AI solution company deeply involved in medical, health insurance, and IoT fields. The advantages of this model are clear: deep industry understanding, high customer stickiness, and strong entry barriers. However, the traditional project model has a natural problem: The more projects one executes, the more personnel are often required. But large models are changing that. Currently, the proportion of recurring revenue from UNISOUNDs enterprise intelligence services has surpassed 60%, with a year-on-year growth of 40% in the scale of repeat purchases. At the same time, the company is deconstructing and standardizing the business capabilities accumulated from past different projects, solidifying them into reusable agent modules. In the past, it was: One client, one demand, one project. Now it is gradually transforming into: Model foundation + standard capability modules + agent orchestration. The key difference lies in that after a project concludes, the capability does not disappear with the projects end; it can continue to solidify and serve as the basis for future deliveries. Thus, the growth logic is beginning to change. In the past, more projects meant more people. In the future, it may transform into: The more projects there are, the more reusable models and agent capabilities become, making the next delivery potentially faster. This could be the second key to understanding the changing value of UNISOUND. It is transitioning from relying on engineering capability to earn money, to gradually shifting towards earning through the reuse of model capabilities and business assets. Healthcare is not just an industry market. Where do these reusable capabilities originate? Healthcare is a significant answer. As of the first half of 2026, UNISOUND has served over 470 medical institutions, more than 80% of which are tertiary hospitals. In the scenario of medical record quality control, AI has managed to compress the review time of a single case to under 10 seconds, improving manual quality control efficiency by 80%; in the Jiangsu health insurance intelligent business handling system, human input has been reduced by approximately 65%. However, if these are merely understood as industry cases, it underestimates the significance of healthcare for a large model company. What makes healthcare truly distinctive is that: It is complex enough and serious enough. Writing a piece of copy may allow for some discrepancies. But medical records, health insurance reviews, and auxiliary diagnoses cannot accept nearly correct. How does an agent decompose tasks? Which steps can be automated? Which nodes must be manually confirmed? What errors absolutely cannot occur? These questions cannot simply be resolved through benchmarks. They must enter the real world. Thus, for UNISOUND, healthcare is transitioning from a past business scenario into a training ground for models and agents. Models enter real scenarios to resolve issues; real scenarios, in turn, assist models in solidifying more complex execution capabilities. This is also where the emphasis on strong foundational models deep application is truly compelling. Application is not the endpoint of the model; it can also be the starting point for the model to continue evolving. Model competition is also beginning to change its yardstick. All commercial changes ultimately return to the model itself. If the underlying model is not strong enough, even the best business model is hard to establish. In June of this year, UNISOUND released the U2 native intelligent agent large model. Its core focus is not merely on pursuing a larger parameter scale but emphasizes the execution of complex tasks and high intelligence density. The logic behind this is quite direct: In the future AI competition, it will not only be about whose model is larger, but rather How much effective intelligence can be produced by a unit of computational power. Especially entering the Agent era, this question will become even more critical. A task may call the model dozens or even hundreds of times in succession, and every inference cost will amplify. Thus, the commercialization of large models ultimately becomes a multiplication problem: Model capability inference efficiency scenario value. Missing any one of these makes it challenging to establish. What is interesting about UNISOUND's interim report is that these variables have begun to move simultaneously for the first time: Model capability enhancement leads to agent execution capability; Agents enter real workflows, resulting in more Token calls; Token calls expand and start converting to high-gross-margin income; Real industry projects continue to solidify, further forming reusable models and business assets. A commercial flywheel is beginning to emerge. This also makes the narrowing loss worth reevaluating. R&D continues to grow rapidly, but commercialization is starting to accelerate in tandem. If this trend continues, for a large model company, scale will no longer just be a cost; it may also become a leverage for revenue and profit. More important than the number of Tokens is the value of Tokens. This raises an even more intriguing question: Should a large model company really be pursuing more Tokens? Not necessarily. If a Token simply completes a simple chat, its commercial value is limited. But if a string of Tokens accomplishes a medical record quality control, a health insurance review, or even drives a complex business process execution, its value is evidently entirely different. Therefore, what the large model industry truly needs to compete in the future may not just be the quantity of Tokens. But rather: What significant problems does each Token solve? This also corresponds to the two concepts that UNISOUND has consistently emphasized: Intelligence density and Token value. The former determines how much intelligence can be generated from a unit of computational power, while the latter defines how much real value these intelligences can ultimately create. The combination of both is what truly reflects an AI company's commercial efficiency. In the past, when the market viewed UNISOUND, it saw medical care, IoT, chips, voice, and various AI projects. In the era of large models, these seemingly fragmented capabilities are beginning to converge anew: Models provide the foundation for intelligence; Real scenarios provide high-value tasks; Agents connect business workflows; Tokens begin to transform intelligence into revenue. As a result, the capabilities accumulated by a company over the past decade are starting to gradually point towards the same commercial logic. The Chinese large model industry has spent three years sharing technology stories. By 2026, the market is starting to care more about another matter: Can technology deliver? Whose model is truly being used continuously? Who can enter real workflows? Who can turn intelligence into high-gross-margin, repeatable, and sustainable revenue? From this perspective, what truly deserves attention in UNISOUND's interim report is not just that a particular number suddenly looks appealing. But rather that several things, once seen as difficult to align simultaneously, are happening together: Models continue to receive investment, while commercialization begins to accelerate; industry business continues to grow, while US dollar Token revenue starts to steepen; projects are still being delivered but are now solidifying into reusable intelligent assets. In the past, the market viewed UNISOUND more as a company with AI technical capabilities providing industry solutions. Now, this yardstick may need to be adjusted. Because when models begin to directly generate revenue, when one-time deliveries start to transform into continuous usage, and when industry know-how starts converting into reusable assets, a company's value logic will change accordingly. What UNISOUND is truly welcoming may not be a mere gear shift in business. But rather a moment reevaluated by the market.