GuShiio.com Wen| Financial Stories Club, author| Dong Hong, Editor| Wan Tiannan
The wind that the big factory group favors DeepSeek blows hard and hard.
On February 15, National APP WeChat officially tested new AI search functions, connecting DeepSeek-R1 ‘s in-depth thinking service.
Some test users can click on the magnifying glass icon at the top of WeChat to see the words searched by AI, and they can enter the full blood version of DeepSeek-R1 for free; if the entry is not displayed, it means that the user account has not been covered by this gray test, and you can wait patiently for subsequent opening.
With WeChat’s 1 billion-level daily activity, it is expected to lead the public collective into the AI era.
On February 13, Tencent AI assistant Yuan Bao welcomed a major update, announcing that it would support both Hunyuan and DeepSeek models, allowing users to switch between the two at will. Among the many self-developed large-scale model manufacturers, Tencent has thus become the first company to connect DeepSeek to its main product.
In addition, on February 21, Tencent’s enterprise WeChat and tactical competitive mobile game “Peace Elite” also announced access to DeepSeek.
In the past, large factories that were used to having separate hospitals and high walls have embarked on the path of interconnection and inclusiveness of openness in the era of AI models.
1.DeepSeek’s big factory circle of friends has exploded!
DeepSeek is becoming a favorite of major manufacturers, and Tencent is not the only one who extends an olive branch.
On February 16, Baidu announced that its Baidu search and Wenxin Agent platforms will be fully connected to DeepSeek, so that users can ask questions and call DeepSeek at any time within the station.
According to incomplete statistics from “Financial Story Fair”, at present, major manufacturers such as Baidu, Tencent, Byte, Ali, Huawei, and Jingdong have all integrated DeepSeek into their own systems, and the ripples have already spread to automobiles, energy, finance, and cloud., communications and many other business formats.
For a long time, big factories have strict borders and are accustomed to building high walls. Taobao, Baidu, Taobao, WeChat, ByteDance, etc. have also blocked each other.
In the past two years, big factories have successively released their own AI models, but they have also not connected to opponents ‘AI models.
Nowadays, why are major manufacturers willing to break down barriers and pamper DeepSeek together?
First, DeepSeek provides everyone with cheap and easy-to-use shovels in an open source manner, making it easier for giants from all walks of life to create more convenient new tools for themselves, laying the foundation for possible gold mines first.
Subtext: Who doesn’t love good things? Get in the car quickly, maybe you can hit the golden egg first.
Insiders at Baidu told Financial Stories that Baidu, a big model that is easy to use, will open its doors, and DeepSeek’s performance is indeed good.
Second, catch a ride and eat more spilled traffic and opportunities.
Only 20 days after DeepSeek was launched, its daily activity exceeded 20 million, exceeding the daily activity of Doubao by 16.95 million, reaching 40% of ChatGPT.
Since January 27, DeepSeek’s app has always ranked first in the total free download list of Apple’s App Store in China. Its growth curve (only 7 days for 100 million users) is faster than ChatGPT, and also faster than TikTok and Dianduo.
Data source: public information
However, after the explosion, DeepSeek remained Buddhist and did not take measures to alleviate the red lights frequently lit due to busy and crowded users during use. This led to the leakage of surging traffic and demand opportunities on the C-side, which could be absorbed by enterprises with carrying capacity., using super platform (APP)+ advanced models to form a synergy effect of 1+1>2.
Tencent, which embraces DeepSeek most resolutely, is an example of some new customers who were relatively alienated from AI. It may have been when they contacted DeepSeek on WeChat and opened the door to the AI model.
In addition, CITIC Securities likened DeepSeek to the disenchantment of American AI in Shangganling in China’s science and technology field. China AI has established a monument and even brought opportunities for asset revaluation to China companies.
On February 16, Tencent’s share price opened higher by more than 6%, with an intraday increase of more than 7%. It once touched as high as HK$512/share, setting a new high of more than three years, and continued to rise for many days.
From the perspective of the capital market, the Wang Bang combination of social +AI will open up a new track for Tencent ‘s precise service payment, adding its endowments in multiple fields such as e-commerce, payment, and news. The ceiling of ARPU value for single users will be greatly raised, opening up the company’s overall imagination.
CITIC Securities also asserted that China’s AI iteration is accelerating, and domestic large models and intelligent cloud companies are expected to benefit significantly. The Internet super platform is not only an important implementation scenario for AI, but also a supplier of intelligent cloud and large model services, which is expected to bring value revaluation.
Third, strong alliances enrich the user experience.
Subtext: Capture picky users with more powerful products.
“Financial Story Fair” has continuously tried DeepSeek’s in-depth thinking function on WeChat. It is indeed a lot smoother. It has hardly encountered any stuck situations and is also very convenient. There is no need to jump out and open an additional app.
Fourth, the sea embraces all rivers, and tolerance is great.
Subtext: Even the bachelor next door is dancing with DeepSeek. I can’t be looked down on lightly.
The reason why a giant is called a giant is not only because the family is big and rich, but also because the structure is big.
When dealing with DeepSeek, major manufacturers have shown an open and inclusive attitude.
Baidu told “Financial Stories Club” to take the essence of it and recruit wise people if it is easy to use;
Insiders of Tencent confirmed to the Financial Story Fair that it can basically be understood that Tencent is still very open. If the AI models of other major manufacturers or startups perform well, they may be connected to the Tencent ecosystem.
Ali emphasized that Ali has always adopted an open source strategy. rdquo; Alibaba co-founder and chairman of the board of directors Cai Chongxin recently publicly stated that DeepSeek teaches us the value of open source. If your only goal in life today is to develop a closed-source AI system, the brightest doctoral student, Nobel Prize winner, I personally think the value of this effort is approaching zero.” rdquo;
2.“My biological son is a meal, and my godson is a snack?
Although major factories have connected DeepSeek with practical actions, compared with their own son who developed a large model himself, his godson DeepSeek is ultimately a child from another family, more like snacks reserved for users than meals.
Tencent can be said to be the most comprehensive player connected to DeepSeek, including dozens of products such as Tencent Yuanbao, WeChat, ima, Tencent Document, QQ Browser, QQ Music, and Enterprise WeChat, which have all attracted DeepSeek.
Take the most important WeChat as an example. In the previous article,”Financial Story Club” asked several questions continuously to DeepSeek in the WeChat search. It felt that it was indeed slippery and had hardly encountered any problems, but the in-depth experience was very limited.
Perhaps the integration process was relatively hasty, just like putting DeepSeek into the WeChat search system, and the basic functions were not perfect.
For example, no historical records are retained, making it difficult to trace back previous queries. Even after the user quits the chat interface, the current conversation memory content will be directly destroyed; there is no effective linkage between contexts, and every interaction seems to exist in isolation; File upload buttons are missing, which limits usage scenarios; personal data is not connected and cannot provide personalized experiences; relationship chains are not integrated, and social benefits cannot be fully utilized; There is even a lack of synergy with other WeChat functions, just like an isolated island.
Baidu Search claims to be connected to the DeepSeek full-blood version, but the search entrance and functional experience have not changed significantly. The problems in WeChat search also exist. Compared with WeChat search, only DeepSeek’s online search function can be used in Baidu search, lacking user use. It is the most frequent and in-depth thinking button that best reflects its technical advantages.
It is worth noting that AI is used in Baidu search. Without selecting a specific model full blood version of DeepSeek, the answer and interaction is based on Baidu’s technology and algorithms by default.
Source: Baidu Search APP
In other words, in Baidu search, DeepSeek ranks second in the order of users using AI products.
In addition, Baidu itself has a variety of practical functions, such as maps, libraries, encyclopedias, etc., but after connecting to DeepSeek, there is almost no linkage with these functions.
Taking maps as an example, when users search for location-related information, they cannot directly obtain content associated with the location’s navigation and surrounding information on Baidu maps through DeepSeek; when searching for knowledge content, they cannot quickly link to Baidu Library. Relevant documentation data.
The separation between such functions prevents users from obtaining a one-stop service experience.
ByteDance is mainly connected to DeepSeek by Flying Books and Buttons (agent creation platform). After experience, there are also limitations.
Specifically, some users reported that when creating the Flying Book smart assistant, they would give a prompt saying that it could not be used and that the usage limit had been reached, which indicated that users could not order DeepSeek to move bricks without restrictions.
Secondly, Feishu integrates DeepSeek into document collaboration, but the operation path is very hidden. For some novices, it may take some time to find and explore. For example, the author searched for a long time and finally found the entrance in a small + place in the upper right corner of the form.
Button has limited functional expansion. Currently, only DeepSeek functional modules supported by the platform can be used. They cannot freely expand functions like some developers independently integrate methods. For users with complex functional needs, they may not be able to meet their personalized needs.
In short, although major manufacturers have all connected to DeepSeek, the integration and collaboration with their own businesses is not deep enough.
3.Pressure is on the big factory
The charm of scientific and technological competition lies in that each mountain is always higher than the other, and the best will have the best.
DeepSeek’s open-hanging performance, the big manufacturers praised it and moved closer, but the pressure was probably too great.
Even OpenAI, the benchmark in the AI world, was a little discouraged when talking about DeepSeek. When the new o3-mini model was launched in early February, its founder Ultraman rarely admitted his mistake and stood on the wrong side of history (closed source). The adjustment after reflection followed: OpenAI’s upcoming GPT-5 will be released free to all users by then, and new plans will be made on open source.
Meta, which once dominated open source models, has asked its technical team to work overnight in the past few months to replicate the DeepSeek path.
This is true for OpenAI, let alone domestic players who already have a generational gap. After a brief period of panic and judgment, Dachang successively made new statements and new actions.
At the beginning of the year, when a team of the Bytedoubao Model was on vacation overseas, the huge wave caused by DeepSeek attracted Zhang Yiming’s attention, and the team immediately returned to China.
At the latest Byte staff meeting, Liang Rubo said that compared with some fast-rising startups, Byte lacks sensitivity in the face of this new AI opportunity, which makes Byte lag behind in the high-end AI competition. rdquo;
The implication is that Byte missed some key technology nodes.
Baidu has taken a big turn in direction, from closed source to open source. Robin Li, chairman of Baidu, revealed that Wenxin Model 4.5 will be open source, and 4.5 will be the most powerful model in Baidu’s history. He said that the decision of the Open Source 4.5 series stems from confidence in technological leadership. Open Source will further promote the widespread application of the Wenxin Model and expand its influence in more scenarios, but what I want to emphasize is that regardless of Open Source, the basic model has real value only when solving real-world problems on a large scale.
In addition, Baidu will accelerate the performance upgrade and cost reduction of the Wenxin model. Wenxin Yiyan will be fully free from April 1.
Ali continues to strengthen its open source route. Its Tongyi Thousand Questions Model is now the world’s largest open source model group, with more than 90,000 derivative models, ranking first in the world, surpassing the LLAMA open source model developed by Meta. Even DeepSeek is trained based on the Tongyi Thousand Questions Basic Model.
In this AI horse racing event, Tencent took the most vigorous action. Not only did the whole family access DeepSeek in buckets, but it also became the first company to connect DeepSeek to its own model.
Regarding the conflicts that may lead to mutual interaction between left and right hands, or whether Tencent will relax its own model research and development, according to the “Alphabet List” report, Tencent has made it clear that it will not give up its self-research strategy. In the future, Tencent will still stick to it in self-research on large models. rdquo;
“Financial Story Fair” cautiously believes that giants with a wide range of business scenarios such as Tencent need to have more complete and powerful multimodal capabilities, and this is precisely DeepSeek’s shortcoming.
Tencent will not reduce its investment in self-developed base models. On the contrary, it is more likely to use this as a spur to increase its resource tilt for AI and produce more hard currency with leading indicators with higher efficiency.
Recently, Tencent Hunyuan has begun a small-scale public beta of its long-term thinking model Hunyuan T1, further demonstrating the giant ‘s determination to be self-reliant. zwj;
In short, there are actually no losers in the new ecosystem favored by the big factories.
