Author: Zhang Yongyi
On June 10, the last exam of the 2025 college entrance examination concluded. However, for millions of families of candidates across the country, the upcoming college application process is another kind of college entrance examination.
Faced with admission information from over 3000 universities and aspirations for future lives, candidates and parents find their considerations becoming more concrete. The market has also begun to mass-produce various types of application products and services to meet the personalized needs of post-05 students regarding college applications.
From the planning services that easily exceed tens of thousands of yuan from 'Zhang Xuefeng' to the 'low-end' tutoring priced in the thousands on social platforms, and to the 'AI software' priced in the hundreds, the commercialization wave of college application is gradually reaching its peak.
Many AI products are beginning to break down the information wall that stands before candidates, thanks to their capabilities in data and large models. The emergence of large models in the college application scene allows more candidates and parents to achieve information equality.
As a platform that has been deeply engaged in college entrance examination information services for seven consecutive years, Quark stands behind candidates once again this year—not only releasing the industry's first large model for college applications and a knowledge base but also launching AI-centric features such as 'application reports' and 'deep search for college entrance examination.'
Quark's goal is clear: to answer every open-ended question related to college application through product innovation and AI technology. At the same time, it aims to provide each candidate with a professional application report to assist them in making life decisions.
01 How to create a good 'personalized' application report?
In real-life college application services like Zhang Xuefeng's, teachers must first master a large amount of data and exclusive information to build their own moat. Then, through one-on-one questioning, they can gain a deep understanding of the candidate's information, interests, and family situation for analysis and judgment. After modifications and comparisons, the 'application report' is delivered as the final result to the candidate, who is the user.
In the face of the highly information-dense and long decision-making chain of application reports, what would an AI-generated 'application report' look like?
After Quark launched the 'application report,' I personally experienced this feature. Taking a student from Beijing with a score of 630 in science as an example, who expressed a desire to study law and become a lawyer, I first filled in his personal information and preferences, completing the creation of a personal profile through 12 questions.
After clicking confirm, Quark will begin preparing the report, with a total time of 5-10 minutes and a page count of 15-20 pages.
In this process, Quark will rely on the large college application model to provide candidates with personalized planning suggestions through Agent calls. Finally, it outputs three different reports with varied focuses (major priority, institution priority, and regional priority), with the report contents including strategy design, detailed institution and major information, and application form interpretation, which users can directly add to their application forms or export as PDF.
From the results of the report, it can understand my preference for law and develop a gradient planning combining 985 and 211 universities and specialized subjects.
It can also analyze the city, tuition fees, and employment situations of majors, integrating this information into the recommended schools and majors, allowing candidates to grasp all the information more clearly rather than just simple university and major information.
Data shows that every year only 2% of candidates choose to seek offline consultation. Therefore, for the 98% of candidates, Quark's emergence allows more candidates to no longer be troubled by regional and cost issues, and the information gap in college applications is becoming smaller and smaller.
Moreover, with the support of the deep search function for the college entrance examination, even for some open-ended and highly colloquial descriptions, Quark can provide more realistic reference suggestions.
For example, I used the prompt 'Shandong science student, 647 points, recommend possible 985 universities, can collaborate with foreign institutions, wants to pursue graduate studies or study abroad, do some application reporting' to experience the deep search capability for the college entrance examination.
Under this prompt, Quark will first analyze the candidate's core demands—score, subject selection, interests, and regional preferences—then conduct multidimensional matching and reasoning within its vast knowledge base on college entrance examination. This knowledge base not only contains historical admission data, major information, employment rates, and further study rates, but also integrates a wealth of unstructured knowledge about industry development trends and the connections between majors and professions.
02 How the 'expert brain' is formed
To ensure that the user experience at the front end is more accurate, Quark has chosen to further invest in the expansion of AI model capabilities this year, with a 100-fold increase in computing power.
Although based on the Universal Question Model, Quark's college application model is not merely a fine-tuning of the general model, but rather a refined mechanism guided by real-life evaluations from college application experts to enable the model to truly 'think like application experts and provide suggestions.'
To achieve this goal, Quark first needs to teach AI to mimic the 'thinking chain' of real-life experts. During the instruction fine-tuning phase, the R&D team structured the multi-round real conversations between hundreds of experienced college application planners and candidates, distilling complete analytical paths and communication styles. These high-quality supervised data, containing tens of thousands of real expert 'reasoning chains,' serve as the 'textbook' for the large model to learn the analysis process of human experts.
In this regard, Quark demonstrates its core advantage. 'Quark's data comes from authoritative materials released by official examination authorities, similar to the industry-recognized 'big thick book',' emphasized Mr. Ren, an expert involved in training Quark's AI application model. This stands in sharp contrast to many large models that rely on unverified old data scraped from the internet, fundamentally eliminating absurd AI illusions such as 'students scoring 500 being recommended for 985 universities,' ensuring the accuracy and authority of the recommendations.
03 The pulse of the times
From 'How the AI college application large model is formed,' you will find that a large model rooted in specific application scenarios can produce accurate results for candidates, thanks to the help of real college application experts.
Every year during the college entrance examination season, we see more professional application counseling experts like Zhang Xuefeng, but in reality, the space is still filled with numerous 'application mentors' whose quality is hard to guarantee, targeting anxious students and parents.
The reason these services can still thrive every year is that the significance of the application process has far surpassed 'choosing schools and majors'; it has evolved into the 'first career planning' involving the entire family. It can even be said that it has become the 'pulse of the times' for tens of millions of candidates each year.
However, what Zhang Xuefeng represents is the expensive solution at the top of the pyramid. One Zhang Xuefeng has limited energy; thus, his service is destined to be a 'luxury' for a few. Behind him lies a much larger and mixed market, where countless institutions and individuals claiming to be 'experts' provide services of questionable quality, reaping benefits from ordinary families who are equally anxious but unable to access top resources.
If Zhang Xuefeng's core barrier is his personal experience and information accumulation, then Quark's approach is to internalize the decision-making logic and experience of hundreds of senior planners through the large college application model, combined with the largest, real-time updated knowledge base on college entrance examination in the country, attempting to transform the past reliance on personal, expensive, and non-standard 'expert services' into a standardized, high-quality 'AI advisor' accessible to every ordinary person for free.
Some say that the AI college application tool launched by Quark is 'going to flip the table,' but that is not entirely accurate. It is not flipping the table—it's changing to a bigger table, allowing more people to sit at it. This table does not require reservations or consultation fees. As long as you can open your phone and fill out a profile, it can provide you with a truly logical and data-driven application report.
At this table, candidates from Liangshan and Hangzhou see the same report structure, the same professional dimensions, and the same recommendation logic. AI has brought their starting points a little closer. The college entrance examination is an opportunity to change one's destiny, and the significance of technology is to make that 'opportunity' a little fairer. Such opportunities should never be placed solely on a VIP gold card table.
The user data publicly disclosed by Quark at the June 12 press conference serves as the best footnote for this—its college entrance examination service has helped a total of 120 million users, with users from third-tier cities and below accounting for over 50%.