GuShiio.com 
Image source: Generated by AI
At the beginning of 2025, there has been a wave of “reasoning fever” in the big model industry. Now, AI searches that cannot think deeply are no longer qualified.
Recently, Quark AI search launched “Deep Thinking”, and an AI application that uses self-developed reasoning on large models has arrived!
The emergence of AI search has solved complex problems that required multiple searches in the past.
For example, if AI can help you analyze the trend of gold prices, you only need to throw it to the AI to search for a question, and it can help you integrate the judgments of Goldman Sachs, UBS, JPMorgan Chase and other institutions into a table for its judgment. Key factors such as monetary policy and supply and demand structure can be presented together.
In addition to dealing with study and work needs, AI search can also serve as a “life assistant” for users. Whether it is asking about health questions such as “What medicine should I take for headaches and fever” or educational consultations such as “What should I prepare for the beginning of my career?”, matched with in-depth thinking skills, the answers provided by AI search are comprehensive and reliable.
When deep thinking and AI search are deeply integrated, the search industry is entering a new era of AI search.
The user experience is being innovated. From the “aggregated information” of traditional search engines to the “solutions” of the AI era, AI search will also lead the entire AI application into the 2.0 stage.
AI Search Enter
The match-point moment for “deep thinking”
During the Spring Festival, DeepSeek once surpassed ChatGPT and ranked first in downloads on Apple’s free app list. Behind this, DeepSeek R1 relies on its deep thinking ability to bring comprehensive innovation to the user experience.
The performance of “deep thinking” has become a new criterion for testing whether an AI application is advanced.
Since February, news of AI search testing water testing and deep thinking function has become increasingly frequent:
On February 16, Tencent announced that it would connect DeepSeek-R1 to Tencent Yuanbao, QQ browser and other products, and launch a deep thinking mode in the QQ browser; Xiaohongshu’s AI search product “Diandian” is testing the deep thinking function, and the one that was exposed to it was also DeepSeek R1; At the end of February, Ali AI To C’s flagship Quark also embedded “deep thinking” into the AI search box. Whether it is on the Quark App or Quark PC,”deep thinking” not only provides users with answers, but also analyzes and thinks., make plans.
Why are all searches focused on “deep thinking”?This starts with the last round of AI search.
Although compared to traditional search engines, AI search has optimized the user experience and can automatically help users find and integrate the most relevant content from the entire network information. However, due to AI’s lack of in-depth thinking ability, it is still difficult to fully understand user needs in some scenarios, resulting in average accuracy of content generated by previous generation AI searches.
For example, Google AI search had just been launched, and it apologized for finding the answer “Suggest that pizza should be coated with water and let humans eat stones.” This is because it used the wrong information source during the search process.
After linking the “deep thinking” ability, the advanced version of AI search can not only clarify user intentions and understand user’s “subtext” based on questions, but also specifically screen information on the entire network based on these needs, and then based on thinking, integrate the most suitable content.
Regarding the question of searching for health content “recipes of the week suitable for office workers to lose fat”, we first asked Quark.
In response to this seemingly simple question, Quark spent 5 seconds finalizing the key formulation idea: it took into account the keyword “office worker” and not only clarified that the recipe should be suitable for office workers ‘schedules, but also provided a set of “Fat-reducing Diet Design Principles” determine the hard conditions that each meal needs to meet.
Based on the above principles, Quark freely integrated the previously collected web information and gave a recipe arrangement for seven days a week.
In addition to the requirements of “one week’s recipes”, Quark AI Search also carefully provides key daily precautions based on users ‘fat-reduction needs. It is more like a “close secretary” who will draw inferences from one example and think carefully.
It can be seen that after being equipped with the in-depth thinking model, AI search comprehensively innovates the user’s search experience. It can not only give good answers that provide inspiration and inspiration sources, but also get a good solution.
Deep thinking and AI search
A golden couple
With “deep thinking” becoming standard, AI search products have taken different development paths:
Currently, AI search + in-depth thinking on the market can be divided into two categories:
The first is to connect the DeepSeek model for in-depth thinking, such as Little Red Book “Diandian”;
The second is the self-research faction, such as Ali’s Quark.
Separating from the product form, one such as DeepSeek and Tencent Yuanbao has directly implanted “online search” capabilities on their own large model platforms. The other is like Quark, adding in-depth thinking based on years of deep AI search.
No matter which of the above, it seems that they all reach the same goal. 2.0 The version of AI search uses the deep thinking model as an “external brain”, and AI search also happens to be the golden partner of “deep thinking”:
When the large model quotes fabricated documents and virtual news clearly in the illusion, AI search can ensure the authenticity and reliability of the information source through real-time networking and significantly reduce the illusion problem of the large model. In the deep thinking mode, the advantage of AI search is traceability. It clearly lists every source of information it uses to ensure the authenticity of the information.
Among the diverse AI search products,What determines the experience of AI search + deep thinking?After light cone intelligence evaluated a number of AI search products on the market,Data accumulation and modeling capabilities have become two key indicators.
First of all, doing AI search requires strong high-quality content support.
In addition to linking high-quality content and historical databases across the network, on many issues involving timeliness, there must be enough new content support to ensure the accuracy and real-time nature of AI answers.
For example, Light Cone Intelligence used “The latest news of Xiaomi Su 7 Ultra” to inquire about multiple products. Among them, the content information given by some models seems to lag behind. In the in-depth thinking process demonstrated, it can be seen that the latest data source found by the model is still a few days ago.
The lagging information source directly leads to the lack of timeliness of the results. During testing, some models could only inform users that Xiaomi Ultra would start delivery within 3 months. The Quark AI search gave a clear answer, that is, delivery will start on March 1.
In the process of thinking, Quark AI directly and clearly marked the information of that day (March 1)
In addition, only with sufficient information sources can AI give full play to its thinking ability and select and summarize more complete answers based on the logic of in-depth thinking.
During the test, when the light cone intelligently asked,”How much water is most appropriate to drink every day? How to determine if you are short of water?” At the time, Quark AI Search quoted 25 search messages, covering health-specific websites, major news platforms and official agency responses, ensuring the comprehensive and professional nature of information sources.
In addition to the accumulation of high-quality content, model capabilities also determine the characteristics of AI search products.
Take the “deep thinking” of Quark AI search online as an example. It is an inference model developed by Quark based on the Aritongyi Thousand Questions Basic Model. It is good at rapid thinking, credible and time-sensitive. The combination of Ali model and Ali application is superior in terms of stability and accuracy in use.
In some vertical search areas, the effect of Quark AI search is more ideal, which is also based on the exploration that the Quark team has been conducting in specific fields for many years.
After the emergence of large model technology, Quark AI Search placed its first important scene in the health field and launched the “AI Health Assistant”. Based on searching for health content and building a self-built health encyclopedia, Quark has changed the user’s ways. There are many pain points when searching for health content. Challenging industry problems and solving the real needs of users, Quark took the first step early with its AI capabilities.
Overall comparison,At present, products with gene search can better leverage the advantages of deep thinking and achieve the effect of 1+1>2. The inference model capabilities used in deep thinking can bring different experiences to the effect of AI search.
From AI search to all-round assistant
AI is everywhere
AI search superimposed with “deep thinking” expands users and industries ‘understanding and scope of search.
Observe a variety of AI search products on the market,It is not difficult to see that AI search is evolving in the direction of “AI all-round assistant”. In addition to ensuring the most basic search functions, it is becoming a functional collection of AI search + everything.
Among them, AI PPT, AI graphics generation, etc. are functions that are in large demand by users. Before the advent of large model technology, it was almost impossible for users to obtain a PPT and picture that could meet their personalized needs, so there would be needs to find templates and pictures. But the experience is neither complete nor efficient.
Compared with other products, Quark AI Search has the most thorough understanding of search requirements and the strongest service capabilities. In terms of AI capabilities, it takes all the AI functions that office workers and student parties can think of: if you need to report, AI PPT automatically generates beautiful documents based on the theme; when writing manuscripts, office workers can directly input the theme, genre, number of words, style, etc. Requirements, you can get a very high-quality manuscript; students doing the questions can also use AI to search for the questions to find solutions, and AI can also help you explain the questions until you learn this kind of problem…
Moreover, Quark also chose to provide all these functions for free. Even the current popular charging items AI PPT, AI resumes, AI life maps, etc. can be used indefinitely, directly saving a membership fee.
This “function collection” model is making AI search a product comparable to the “AI workbench” category and an AI all-round assistant that improves work and learning efficiency.
In the past, our work required the use of multiple products such as search, document writing, and drawing, and multiple interfaces were frequently switched. However, AI search that integrates multiple AI functions can greatly reduce the occurrence of this problem.
In terms of product design, AI applications have begun to understand users and needs better. By identifying and understanding user intentions, it can “deliver” corresponding AI functions that better match the user’s current needs to your eyes, truly assisting users in depth. Take Quark as an example. When you enter questions related to writing in the interface, such as “Help me analyze a company’s business model”, it will automatically jump the content generated by AI search to the AI writing interface for users to modify and adjust.
It can be seen that with the support of AI, AI search products are providing users with more and deeper value, and are growing from a towering tree to a tropical rainforest.
Technological innovation drives the times forward.
In 1998, when people were still struggling to find the information they needed in the early generation of search engines classified according to catalogs, the two founders of Google used the PageRank algorithm to achieve accurate sorting of search results, and the browser represented by Google became the unquestionable king of the market.
Nowadays, with the help of the capabilities of the AI model, people have finally seen new possibilities in search engines that have been developing for nearly 30 years. Compared with traditional browsing engines that are full of advertisements and require users to select and screen, AI search represented by “no advertising” is undoubtedly the next generation of traffic portals.
“AI’s complete subversion of the search industry has already begun.”
