Zhihu Inc.

Ladies and gentlemen, thank you for standing by and welcome to the Zhihu Inc. Second Quarter 2026 Financial Results Conference Call. At this time, all participants are in a listen-only mode. After the speaker's presentation, there'll be a question and answer session. Today's conference is being recorded and webcasted. At this time, I would like to turn the conference over to Demi Leung of Investor Relations. Please go ahead, ma'am.

Thank you, Sharon.

Hello, everyone. Welcome to Zhihu Second Quarter 2026 Financial Results Conference Call. Joining me today from our senior management team are Mr. Zhou Yuan, Founder, Chairman, Security Officer, Mr. Wang Han, Chief Financial Officer, and Mr. Zhang Longle, Chief Offering Officer. Before we begin, I'd like to remind you that today's discussion will include forward-looking statements made under the safe harbor provisions of the U.S. private security Legislation Reform Act of 1995. These statements involve inherent risks and uncertainties. As such, actual results may be materially different from the views expressed today. Further information regarding this is included in our public filings with the U.S. Security and Exchange Commission and the Hong Kong Stock Exchange. The company does not assume any obligation to update any forward-looking statements, except as required under applicable law. Additionally, today's discussion will include both GAAP and non-GAAP financial measures for comparison purposes only. For a reconciliation of these non-GAAP measures to the most directly comparable GAAP measures, please refer to our earnings release issued earlier today. of this conference call will be available on our IR website at irdoctorwho.com. And today, Mr. Zhou Yuan will deliver prepared remarks in Chinese and followed by English translation. Please go ahead, sir.

Hello, everyone. I'm Zhou Yuan. Thank you for attending our second quarter conference. Today, I'm going to talk about three aspects. First, Hello, everyone, and thank you for joining Zhihu's second quarter 2026 earnings call.

Today, I will cover three areas. Our core business performance in the second quarter, how AI is expanding the value of our community, content, IT, and expert network, and our priorities for...

In general, the core business is stable. Some businesses are still in the process of adjustment and recovery. In the second quarter, the company's total income was RMB6.9 billion, a decline of 3.7%, and a growth of 5.9%. The income has improved significantly, and has also been further narrowed. Due to the change in the income structure and the stage-based business investment, the net loss was RMB1025 million. In the second quarter, the core users' daily usage time is 39 minutes, which is the same as last year. The content of the new high-quality content of the community has increased by more than 16% in the same ratio, extending the previous high-quality content of the core users. Let me start with our core businesses.

Overall, our core business remains stable, while some areas continue to adjust and recover. In the second quarter, total revenue was RMB $619 million, down 3.7% year-over-year and up 5.9% sequentially, with a year-over-year decline narrowing further. Adjusted net loss was RMB $10.25 million, reflecting changes in revenue mix and phased business investments. Our CFO will discuss the financial details shortly, Community engagement among core users remains stable. Average daily time spent was about 39 minutes. Broadly, daily creation of high-quality content grew more than 16%. As AIGC makes content creation easier, authentic experiences, clear sourcing, and professional judgment are becoming more valuable. These remain the foundation of Zhihu and our long-term value.

In the second quarter, the revenue was RMB1.99 billion, which fell by 10.7% and increased by 4%. In this quarter, we continue to focus on technology, vehicles, 3C home appliances, and games, which are in the industry that are more compatible with smart users and content advantages. Our performance-related products consumed 31% of the overall consumption, and Yu Xie. Now, the growth of the product has not completely transformed into the recovery of the overall sales revenue, but it has proven that we continue to focus on the direction of focusing on the focus industry to strengthen the product and the ability to calculate. The sales industry is still in a structural recovery stage. Next, we will continue to improve the professional content, user decision scenarios,

Turning to marketing services, revenue will be 199 million, down 10.7% year-over-year and up 4% sequentially. We continue to focus on key verticals including technology, automotive, consumer electronics and home appliances, and gaming. Spending or performance 1% sequentially including a 22% increase in gaming. This has not yet translated into a broader recovery in the marketing revenue, but it supports our strategy of strengthening key verticals, products, and algorithms. Marketing services remain in structural recovery and will continue to improve the matching of professional content, user needs, and client demand.

We will continue to optimize the experience of member content and products, and expand the content of knowledge to promote the multi-dimensional structure of non-conventional formation of stories plus knowledge. In this quarter, IP operations continue to contribute to increase. In the second quarter, the number of copyright cooperatives increased by 105% and increased by 671%. The IP from single-party copyright sales to multi-format full-life cycle development and extension. Overall, the challenges we face still exist. For paid content and IP, revenue reached RMB 426 million, up 4.4% year-over-year and 5.9% sequentially. Average monthly subscribing members were

13.11 million, broadly stable, while member ARPU also remains stable. We'll continue to improve membership content and product experience while expanding knowledge-based offerings. IP operations continue to contribute incremental growth, licensing increasing 105% sequentially and 600% year-over-year. We're expanding Yanyan Story IP from one-time licensing toward multi-format, for life cycle development. Overall, challenges remain, but our business mix continues to adjust, and our strategic direction remains unchanged. We'll continue to stabilize our core businesses while maintaining the investment discipline.

Next, I'd like to talk about the importance of AI-based calls. Calls come from the community, but they're not limited to the community. The value of long-term accumulation in the community is used by more people through new products, content and application scenarios. Real users, professional content, IP and expert networks are still our most important assets. AI is helping these capabilities to enter more new applications, content and business scenarios. First, the community will become more open. AI is changing the way of discovery, organization, and use. For consumers, opening up does not mean changing the real, professional, and possible community positioning. It is not simply moving the content of the community to the outside, but allowing high-quality content and professional creators to be discovered by more users and developers through more tools and AI applications. This kind of opening up has clear boundaries. In the past, users mainly recommended the discovery of information flow through search. Today, we are promoting the further integration of call-to-answer and call-to-search, so that AI search becomes the structural entrance to the discovery of social content, rather than replacing the original content. Our initial product test shows that this work has had a positive impact on the user's workflow. We are still observing its usage range and long-term effects. At the same time, we are also exploring community intelligence. AI Mountain View is one of the specific attempts. We hope AI Mountain View can become a partner to help users explore and call out the community. to make it more convenient for users to discover real people, professional content, and different perspectives. At present, AI Look 3 is still in the early stage of testing. We will focus on observing its actual driving of AI discovery, user interaction, and flow, rather than simply pursuing the use of scale. As I mentioned before, AI Works has continued to advance so far. The platform has loaded more than 2,600 AI projects, The open data platform has already exceeded 17,600 users, 20% of which are new developers other than call creators. At the same time, we also released a new version of call CLI, further improving the convenience of developers and users to find and use call content in different AI tools. What does this mean? This means that the creator and professional ability of the long-term collection of call content will not only be found and used in the Zhihu app. Through AI search, AI open platform and development tools, they can enter more products and use scenarios, users, developers, and customers. We think that AI has become a new medium for rapid change. A more open Zhihu can let the存量内容被重新发现, Next, I would like to focus on Zhihu in the AR area. Zhihu starts with the community, means bringing the value accumulated within Zhihu to more people through new products, formats, and use cases.

A real user's professional content IP and our next expert network remain our most important assets. AI is helping extend these capabilities into new applications and commercial scenarios. First, the community will become more open. AI is changing how content is discovered and used. For Zhihu, openness does not mean changing our positioning. It means enabling the high quality content to reach more users and developers through the AI tools and applications. This requires clear cultures around content sourcing, creator rights, and community values. We are further integrating Zhuhu Zhuda with Zhuhu Search, making AI-powered search a new gateway to community content rather than a substitute for it. Earlier testing has shown positive signals for user retention. We're also exploring community agents. AI Kunshan is one example. Rather than another general-purpose chatbot, we wanted to help users explore Zhihu and discover real-people professional content at diverse perspectives. The product remains at an early testing stage, and our focus is on content discovery, interaction, and retention. AI Works now hosts more than 2,600 AI projects. While the open data platform API has attracted more than 17,600 professional developers, about 20% were not previously Zuhu creators. We also released and updated Zuhu CLI, making it easier to discover and use Zuhu content across AI tools. Through AI Search, agents open platforms and developer tools Zhihu's content and professional capabilities can reach more users, developers, and clients beyond the Zhihu app. We believe AI is becoming a new medium that can help existing content reach new audiences, enable new applications, and create new opportunities in content licensing, brand content assets, and expert services.

Now there are a lot of things that happen in the AI environment. So the brand not only needs to be seen by users, but also starts to pay more attention to their own professional information, whether it can be accurately understood, quoted and continuously spread by AI. In order to meet this need, we are helping brand customers to identify content assets with clear sources and professional information. We think Professional content assets are becoming a new brand asset in the era of AI. Its core value is not only to complete this degree, but to help the brand to continue to accumulate affordable content. Let the content invested in the first time be able to generate longer and wider value. It also makes brand information easier to be accurately understood and applied by AI. For this demand, we are constantly trying and promoting AI content asset business. The related business is still in the early stage. In the second quarter, the number of partners increased by 50%. Although today the increase in the number of projects and cooperation finance does not mean that it has formed a stable and large-scale contribution, but it makes us very clear to see the new commercial value of professional content assets in the AI era. Second, marketing services are evolving from traffic value toward the content asset value.

As more information discovery happens through AI, brands increasingly care not only about visibility, but also whether the professional information can be accurately understood and cited by AI. While helping brands build professional content assets with clear sourcing. Unlike a one-off campaign, these assets can continue to be discovered and used across search, AI, and other channels, creating longer-term value. Our AI content asset offering remains at early stage. In the second quarter, the number of clients increased 50% sequentially. This does not yet represent a stable or scalable revenue contribution. but it's beginning to demonstrate new commercial value. Going forward, we'll focus on client outcomes with demand and product standardization.

第三 AI正在推動內容IP進一步的多媒體化。 AI所帶來的另一項變化是進一步降低了優質IP從文字向多媒體形態轉化的門檻。 對於之後來說,這就意味著多年所積累的大量的 The original IP can be more efficient and reachable through slow-moving, short-term vision, and more new content patterns. In the slow-moving project of commercialization realized in this quarter, a considerable part of it comes from the high-quality original IP from 2025 and the previous generation. This means that high-quality content will not lose value because of time. In the first half of the year, we also made some attempts in terms of autonomy and integrity. In the first half of the year, we also made some attempts in terms of autonomy and integrity. In the first half of the year, we also made some attempts in terms of autonomy and integrity. Third, AI is accelerating the expansion of IP into multimedia formats.

AI is lowering the cost and barriers of turning text-based multimedia formats. For Juhu, this allows our large space of original content to be developed more efficiently into comic dramas, short dramas, film and television, and other formats. A meaningful portion of the comic drama projects monetized this quarter came from the works created in 2025 or earlier. These shows that high quality IP can return value over time. What technology improves the development efficiency and expense is rich. In the first half, we also explore in-house with some projects showing encouraging early results. However, we'll remain disinclined and will not materially increase asset-heavy investment. We'll flexibly choose more licensing In-house and commission production based on the project economics. Going forward, we'll continue to strengthen our capabilities in IP selection, development, and cross-format operations to unlock more value from our content library.

第四呢,這個專家數據的解決方案正在從項目的交付走向能力的附用。 through answering questions and creating content, and sharing knowledge and experience with users. So now, we are starting to explore how to use training data, model evaluation, and complex task design to further transform these professional knowledge, experience, and judgment into model power. We have designated this business as a data research laboratory driven by research. Its core is not traditional data markers, but continuous recognition of the ability of models to edit and supplement, and Wei Yao to develop training data and environment to improve the performance of the model. In the first half of the year, we have completed the project in many aspects such as code searching, in-depth research, visual reasoning, and intelligence. And gradually formed a integrated ability to analyze data development and production from the ability of the model to quality inspection and model effect assessment. The long-term collection of expert networks has formed the definition of professional tasks for us to establish quality standards and verify the quality of data. These works provide important mechanisms. The value of experts is not just about participating in the production of data. More importantly, it helps to define professional tasks, establish quality standards, and judge whether model output really meets professional requirements. At present, we are more active in studying the ability of different models to automatically develop the corresponding training data and evaluation program, and to verify the value of data through the actual model effect. We hope that the current data is not just a single exchange, but a set of data. We focus on verifying whether this set of ability can serve different customers, abilities, and models. And the next step is to develop data products, model evaluation, and complex environment solutions.

Fourth, expert data solutions are evolving from project delivery toward reusable capabilities. AI is creating new ways to use Zhuhu's expert capabilities, not only creating content for users. Experts can now help translate the professional knowledge and judgment into model capabilities through training data, model evaluation, and complex task design. We position as a research-driven data lab. We identify model capability gaps and develop training data, evaluation systems, and complex task environments to help improve the model performance. In the first half, we completed projects across coding, search, and deep research, visual reasoning, and agents, while building capabilities spanning model analysis Our expert network plays an important role in defining the professional tasks, setting quality standards, and evaluating whether model outputs meet the professional requirements. We're also becoming more proactive in identifying model gaps, developing training data and evaluation methods in-house, and validating the values through the actual model performance. Our goal is not one of data delivery, but reusable capabilities that can evolve across clients and model iterations. In the next phase, we'll continue this and extend them into products, complex tasks, and environments.

In the second phase, we'll continue this and extend them into products, complex tasks, and environments. In the second phase, we'll continue this and extend them into products, complex tasks, and environments. Finally, let me briefly summarize. In the second quarter, our core community remained resilient, paid content remained stable,

Our IP operation continues to grow and marketing services improve sequentially. Our position in the AI area is also becoming clear. Our authentic, professional, and trustworthy community remain the foundation, while AI helps us reach new users' applications and commercial scenarios.

In the next three and a half years, we will focus on two aspects of our work. One is to continue to consolidate core business. We will continue to verify the new opportunities brought by AI, including a more open community environment, AI-based assets, IP, and expert data. We will pay more attention to the real needs, customer value, and their resale performance, as well as our investment returns. and others. We also clearly see that the initial verification of the new direction does not mean that it has formed a stable growth. Therefore, we will remain cautious in investment and adhere to strict requirements of investment and return. In this process, the profit performance of three seasons may still be affected by the business rhythm and stage investment. In the second half, we'll focus on two priorities.

First, we'll improve user experience, maintaining a stable membership base, improving IP development efficiency, and advancing the structural recovery of marketing services. And second, we'll continue to validate AI-driven opportunities across our open community ecosystem, AI content assets, IP and expert data solution by focusing on real demand, client value, capability reuse, and ROI. Initial validation does not mean these initiatives have become stable growth drivers. Profitability may fluctuate due to the business timing and face investments. Our long-term goals of improving operating efficiency and returning sustainable profitability remain unchanged. Over the long term, we want Ji Hu to be not only an authentic, professional, and trustworthy community, but also an important platform connecting content, knowledge, and AI applications. Thank you. I will now hand over to our CFO to review the quarter's financial performance.

Hi, I will now go over our second quarter issue earlier today. During the second quarter, our revenue trend continued to improve sequentially. supported by growth in paid content and IP operations. At the same time, disciplined cost management drove year-over-year reductions in operating expenses and operating losses. Now, turning to the financial highlights of the second quarter. Our total revenues for the quarter were RMB 690.1 million, down 3.7% year-over-year and up 5.9% sequentially. The sequential growth was primarily driven by content and IP operations. Marketing Services Revenue Award was RMB 199 million compared with RMB 222.8 million in the same period of 2025. Decrease primarily reflected our proactive and ongoing refinement of service offerings. Sequentially, marketing services revenue increased by 4% with increasing traction in key verticals and performance-based products. We also continued to make early progress in AI-related commercial use cases. Content and IPO operations revenue was RMB 425.9 million, up 4.4% year-over-year and 5.9% sequentially, primarily driven by continued growth in IPO operations. Average monthly subscribing members remained stable at 13.1 million. We will continue to strengthen our paid content offerings while developing and modernizing selected IP across multiple formats with disciplined attention to project returns and risk. Our revenues were RMB 86 million in the same period of 2005. The decrease was primarily due to the continuous strategic refinements of our vocational training business.

Sequentially, auto revenues increased by 12.7%, and year-over-year decline continued to narrow.

Our gross profit per quarter was RMB 393.4 million, compared with RMB 448.2 million in the same period of 2005. Gross margin was 57%, compared with 62.5% in the same period of 2005. Decline in gross margin primarily reflected our continued efforts to broaden and enhance our content offerings. Total operating expenses decreased by 13% to RMB 469.4 million from RMB 539.2 million in the same period of 2005, reflecting continued efficiency improvements across our operations. Selling and marketing expenses decreased by 5.4% to RMB 308.7 million from RMB $326.3 million in the same period . Research and development expenses decreased by 25.4% to RMB $108.6 million from RMB $145.7 million in the same period of 2025, primarily attributable to continued improvements in our research and development efficiency. General administrative expenses decreased by 22.7% to RMB 52 million from RMB 67.3 million in the same period of 2025, primarily attributable to lower personnel related expenses.

Period 2025.

On a non-GAAP basis, adjusted loss from operations narrowed by 32% to RMB 48.7 million from RMB 71.5 million in the same period of 2005. Investment income was RMB 16.4 million compared with RMB 140.8 million in the same period of 2005. The decrease was primarily due to an unrealized gain from the fair value of remeasurement of our investment in a privately-hold Company in the same period of 2005. RMB 37.4 million compared with net income of RMB 72.5 million in the same period of 2005. On a non-GAAP basis, adjusted NILOS was RMB 10.3 million compared with adjusted net income of RMB 91.3 million in the same period of 2005. As of June 30, 2026, We had RMB 4.4 billion in cash and cash rate quotes, term deposits, restricted cash, and short-term investments, maintaining a solid liquidity position to support our own. As of June 13, 2016, we had repurchased an aggregate of 41.3 million Class A auto shares for a total consideration of US $77.9 million on those New York Stock Exchange, and the Stock Exchange of Hong Kong. During the second quarter, we repurchased 6.5 million on Class A ordinary shares for a total consideration of US$7.2 million. Looking ahead, we will continue to balance selective investments in new initiatives with operating efficiency. Though quarterly profitability may be affected by this investment, our long-term objective of improving Thank you.

To ask a question, you will need to press star 1 and 1 on your telephone and wait for your name to be announced. To withdraw your question, please press star 1 and 1 again. In the interest of time, please ask one question each time. If you have any follow-up questions, please go back to the queue. Thank you. We will now go to the first question. One moment, please. and your first question today comes from the line of Thomas Chong from Jefferies. Please go ahead.

So I will translate myself. Thank you for taking my question. So my question is, how should we think the revenue and profit trend in the second half of the year? And under the business adjustment and the business investment, are there any fluctuations between quarters? Thanks.

Let me answer your question. I am Zhou Yuan. From the perspective of the next few years, we think it is not suitable to simply follow the return trend of the second quarter. In fact, it is still not the same. Business services are still in the process of structural repair. It will be affected by customer budget and industry demand changes between seasons. Our focus is still to strengthen the basic ability of production and effect advertising to improve the efficiency and value of unit traffic at the head, rather than to promote growth by increasing advertising add-ons.

This is speaking on behalf of CEO. Looking into the second half, we do not think it's appropriate the sequential improvement seen in the second quarter. And different businesses have different operating rhythms. And marketing services remain in a period of structural recovery, like we mentioned before, and may continue to be affected by changes in client budgets and industrial demand. will strengthen our product R&D and especially the performance advertising capabilities and also increase the value per unit of traffic rather than drive.

The content is basically stable, but the operation of the IP itself will be affected by the flow of the project rhythm. Recent industry regulations and requirements for changes may also affect the progress of some projects.

Paid content remains relatively stable. While IP operations will be influenced by project timing, recent changes in industry standards and filing requirements may also affect the launch timing of certain projects.

At the same time, we're still validating the new capabilities.

like AI content assets and expert data solutions. We have seen like mind and client validation, but it might take time until we see scalable revenue contribution and will also require some near-term investments.

So at this stage, we are more concerned about the stability of core business itself. Can new business form a real sustainable customer value? So at this stage, we are more focused on the stability of our core businesses.

while whether new initiatives can create real client value and reusable staff and ROI. We'll continue to invest prudently. Our long-term goals remain to improve our revenue mix and return to sustainable profitability.

Thank you.

Your next question today.

comes from the line of Vicky Y from Citi. Please go ahead. Thank you. Would management share some color about your view on the AI-generated comic dramas and how should we think of two whose advantages? Thank you.

Hello, everyone. I'm Zhang Zuo. Let me answer this question. Overall, we are very optimistic about the market of AI and MNG. And we judge that this industry is still in a relatively obvious stage of upward development. It is mainly reflected in two aspects. One aspect is that the overall production capacity of AI is still improving rapidly. In the past year, whether it is the consistency of the characters, the performance of the lens, the ability to move, or the overall efficiency, all have very obvious improvements. On the other hand, we observe that more and more... ...in this field, we see that content, creativity, topics, and forms of expression are more and more diverse. So, we believe that AI Manage is gradually forming a new form of content consumption. As the production capacity continues to increase, we believe that the overall quality of content and the acceptance of users will continue to rise.

This is speaking on behalf of COO. We are very positive on the AI comic drama market and believe the industry is in a clear growth trend. On one hand, AI content generation capabilities continue to improve rapidly, including character, consistency, visual quality, motion, and the overall production efficiency. And on the other hand, As more creators enter the market, we're seeing greater diversity in ideas and formats. So AI comic dramas are gradually becoming a new form of content consumption. As the production continues to improve, we expect content quality and user acceptance to rise further.

In this process, we also observe that the key points of competition in the entire industry are also changing. In the early days, everyone was still fighting over who could use AI faster and produce content at a lower cost. But as the AI's overall production capacity became more and more widespread, the competition returned to the content itself. Whether there is a good story, whether there is continuous innovation, good visualizing ability, and continuous development of IPs gradually became the key to competition.

At the same time, the basis of computation is also changing. In the early days, the focus was on who could adopt AI faster and produce content at a lower cost. As AI production becomes more widely available, the real scarcity shifts back to the content itself, like the good stories are and the ability to consistently turn IP into compelling virtual content.

这恰恰是知乎非常有优势的地方 过去几年我们通过言言故事 形成了比较完整的原创故事和社区的一个创作者的生态 知乎的优势不只是只有一批IP And this is where Zhigu has a clear advantage through In-N-Story and other products. We have viewed a broad ecosystem.

of original stories and creators. Our advantage is not simply the size of the library, but our ability to continuously generate the new content and identify the strongest titles supported by a well-established creator. And this helps reduce the trail and error costs in IP development and improve the overall efficiency of our content portfolio.

In the first half of the year, we have seen these advantages are gradually verified in the AI comic market. According to some data from the third-party industry, in the first half of the year, Yanyan Story has become one of the leading copyright parties of the original AI comic of Douyin, and entered the top three of the red-collar copyright list. At the same time, we have also started to try to do self-management and process business. From the result, we have also obtained a very good return rate result. We're already seeing these validated in the AI corner.

According to the third quality data, in the first half of 2026, Yan Yan Story became one of the leading IP providers for native e-iconic dramas on TikTok and ranked among the top three IP providers on Hongkong. We also began the in-house and commission production in the second quarter. and have seen encouraging heat rate so far. As AI creates new media formats, a strong story can be visualized and in this way, AI helps extend the life cycle of quality IP and improve the monetization efficiency of our existing content assets.

From a strategy point of view, our core is to continue to play an important role in the content and creator ecosystem. Yichun Wang, Yichun Wang, Yichun Wang

Our focus is to build on Zhihu's strength in the content and our greater ecosystem, like we mentioned before, while using AI to improve the development efficiency across scripting, production, and distribution. will remain flexible in our business models, including licensing in-house products. Where the market demand and project economics are well-validated, we may selectively move further options in the value chain.

In the long term, we believe that the real core of this market is a combination of several capabilities, including continuous supply of original IPs, Over the long term, we believe Ji Hu Smoke in this market will come from a combination of capabilities, including a sustainable supply of the original IP.

content selection based on real user behavior, a stable creator ecosystem, and a clear license framework, and AI-enabled cross-media IP development. Thank you. Next question, please.

Thank you. Your next question today comes from the line of Shuiqing Zhang from CICC, please go ahead. Thanks management for taking my question.

Can management elaborate a little more on the business model for expert data solutions? And what's Zhihu's advantage in this business? Thank you.

Thank you for your question. Our position is a second lab that examines the ability of models to improve. Simply put, we will study the current boundary of the model, and find the task that the model cannot be completed steadily, and then surround these problems, design training data, complex environments, and benchmark these things. Through model training and evaluation, we can verify whether these data can really improve the ability of the model. Simply put, our goal is not simply to deliver data,

as a data lab focused on improving frontier model capabilities. So simply put, we continuously study where models do have capability gaps, identify the tasks they cannot perform reliably, and then design the training data There are currently some problems with our clients.

Anxin Xiong, Ding Xiang, Jie Jue But there are more and more of us to define the direction of the research, the direction of the model's ability to improve, and to commercialize these abilities. As the project accumulates, we will put a lot of the pipeline of data that can be used, the ability of the environment to build a parallel system, and the research know-how to be built, and then we will serve more customers and different models.

Our business model currently includes the customized R&D projects based on the client's frontier needs while we're also increasing our own research and exploring ways to productize selected capabilities. As we complete more projects, we aim to standardize and reuse our data pipeline's environment-building capabilities have developed evaluation frameworks and know-how across like more clients and different generations of models. And these should improve both scalability and capability reuse.

专家呢,就是之后的专家网络是非常重要的生产要素。 而且呢,我们就是首先肯定不是说都是让专家自己个人去标了, 而是建立一套这个识别模型的这个 Li Xue Gou, and then the main task is to design, and then to evaluate the results of these signals, and then to build the environment, and then to verify the effect, and so on. So, especially in code, agent, search, and deep research, these directions, this is to say, let this expert directly remove it, but it is based on this, it is related to artificial fusion and some of this to do model experiments.

to whose expert network is an important part. At the same time, we are building an end-to-end R&D look from identifying model gaps and designing tasks and training signals to building environments and graders and ultimately validating model performance. In areas such as coding, search, and deep research, data production increasingly rely on the model requirements rather than the large-scale manual work. Experts therefore play a greater role in setting standards, providing the professional judgment, and validating the results.

Sun Li, Moxin Diao Yong, Ji Chu, Yan Fa, and so on. We are looking to see if we can effectively improve our model and establish long-term trust with our customers. We are also looking to see if we have the ability to cross-border

The investment profile of this business will differ from the traditional one. Going forward, more investment will be directed toward compute, model usage, and R&D infrastructure. with the goal of building capabilities that can continuously evolve and be reused. So at this stage, we are focused on three things, whether we can consistently improve the model performance, whether the long-term relationships with the clients and whether our capabilities can be reused across them. So if these metrics continue to validate, Thank you.

That concludes today's Q&A session. At this time I will turn the conference back to Demi for any additional or closing remarks. The conference is now concluded. Thank you for attending today's presentation. You may now disconnect.