AI is sweeping the Agent market, Tencent is competing with ByteDance

Written by
Iris Vance
Updated on:June-13th-2025
Recommendation

The competition between technology giants Tencent and ByteDance in the field of AI is intensifying, and the DeepSeek R1 model upgrade has become a new focus.

Core content:
1. DeepSeek R1 model upgrade, Tencent and ByteDance quickly join
2. DeepSeek R1's improvement and optimization in deep thinking ability
3. Tencent and ByteDance's competitive situation in AI intelligent system development platform

Yang Fangxian
Founder of 53A/Most Valuable Expert of Tencent Cloud (TVP)


After a year of price war, will Volcano Engine throw out another "trump card"?





The long-awaited  DeepSeek R2  model has not yet arrived, but DeepSeek  launched  the upgraded version of  R1 on  May  29  .


Tencent quickly followed suit. In less than a day, Tencent Yuanbao, ima , Sogou input method, QQ  browser and other products were the first to connect to  the latest version of  DeepSeek R1-0528  , claiming that multiple capabilities were comprehensively improved. This is an important upgrade of the DeepSeek R1  model after  months, which shows  how much Tencent values ​​DeepSeek  .


One hour after Tencent’s official announcement, Bytedance’s Volcano Engine also officially announced its connection to  DeepSeek R1-0528 , and at the same time launched  a 50  % discount promotion for new customers.


The upgraded  R1  model has enhanced its deep thinking ability. DeepSeek  claims that it has achieved leading results among all current domestic models in multiple benchmarks such as mathematics, programming, and general logic, and its overall performance is close to other international top models such as  GPT-o3  and  Gemini-2.5-Pro . At the same time, it has been further optimized in terms of hallucination improvement, creative writing, and tool calling.


After DeepSeek  became popular, access to it has become a standard feature of domestic Internet application products. Tencent, ByteDance and Alibaba have launched large models that are comparable to  R1  but this still cannot overshadow  the brilliance of DeepSeek  . In  the comment section of DeepSeek's  overseas social media, the most discussed topic among users is still looking forward to the next generation large model  R2 , " We want  R2 ".


Before  DeepSeek's  updated model came out, domestic cloud vendors were still competing with each other.


Especially Tencent and ByteDance. At the Tencent Cloud  AI  Industry Application Summit on  May  21  , an  important change for Tencent Cloud was to upgrade the large model knowledge base for enterprises to an agent development platform. This is consistent with the purpose of ByteDance's  AI  application development platform  Coze , which is to help enterprises implement native  AI  agent systems.


On  June  11  , the Volcano Engine  2025  Power Conference will also be held. The Doubao large model family will be fully upgraded, and AI intelligence in the fields of finance, automobiles, and medical health  will  be  demonstrated. More importantly, the price war for large models was launched at the Volcano Engine Power Conference last  May  , and the price of large models was reduced to the "centimeter era", triggering a wave of price cuts among domestic large model cloud vendors. Will Volcano Engine throw out another "king bomb" this time?


Competition between the two parties in their respective cloud service markets has become increasingly apparent. Over the past month or so, major automotive industry customers that originally belonged to Tencent Cloud, such as Yuanrong Qixing, BMW, Dongfeng Motor, and FAW Bestune, have reached cooperation with ByteDance's Volcano Engine to promote cooperation in areas ranging from  AI  marketing to the implementation of Doubao large-scale models for human-vehicle interaction.


Will Tencent, which is strategically "accelerating  AI  iteration", follow ByteDance's lead?


1

Tencent targets  Agent


After fully integrating into  the Deepseek  big model and reaping the benefits, what is Tencent  AI  's next strategy?



"We propose four accelerations for the implementation of  AI  - accelerating the innovation of large models, accelerating the application of intelligent entities, accelerating the construction of knowledge bases, and accelerating infrastructure upgrades, to promote  AI  technology into thousands of industries and into everyone's life."  said Tang Daosheng, CEO  of Tencent Cloud and Smart Industries Group , at the Tencent Cloud  AI  Industry Application Summit.



In terms of intelligent agents, Tencent is about to launch an intelligent agent  called  AI  College Entrance Examination Pass. Previously, QQ  Browser launched  an intelligent agent called Qbot  driven by Tencent Hunyuan and DeepSeek  dual models  , which can help users perform tasks such as file format conversion, application updates and downloads.



For the  B  -side, Tencent's big model knowledge engine has also been upgraded to an "agent development platform" to help companies build intelligent systems. Users can let the intelligent agent autonomously break down tasks and plan paths, actively select and call tools, and autonomously develop intelligent agents that can complete different tasks. On the intelligent agent side, ByteDance launched  the AI  ​​application development platform  Coze (Chinese name: button) as early as last year to help  -side companies build their own  AI  applications.



In terms of knowledge base construction, Tencent already had an IMA for the  C  -end , and this time it released  the "Tencent LeXiang Enterprise  AI  Knowledge Base" for the B  -end , with the aim of enabling  AI  to better manage and apply corporate knowledge.



The knowledge base question-and-answer capability is one of the most effective ways that big models can bring to enterprises. The client case cited by Tang Daosheng is the robotics company Ecovacs, which used the LeXiang knowledge base to create an  AI  assistant, "boosting the labor efficiency of stores across the country, achieving double-digit improvements, and saving millions of yuan in annual operating costs."



Creating intelligent entities and building knowledge bases are the current commercialization directions of   AI  applications, and are also Tencent 's obvious move to catch up with ByteDance in AI  applications.

 




On the same day as the Tencent Cloud  AI  Industry Application Summit, Bytedance’s Feishu also launched “Feishu Knowledge Q&A”, which is essentially  an AI  knowledge base tool that integrates  the DeepSeek R1  full-blooded model. Users can access documents, messages, files and other information within their company through question-and-answer methods to generate more accurate and reliable answers in work scenarios.



 Five  days later , Enterprise WeChat upgraded its "intelligent robot", and its new capabilities matched the functions of "Feishu Knowledge Q&A".



Enterprise WeChat users can find answers to their corporate work through "intelligent robots". The principle is also based on internal information uploaded by the company, combined with deep thinking of large model capabilities such as  DeepSeek  and Hunyuan, and accurately respond to employees' various work-related questions. It is said to provide support for adding online documents, Weipan folders and other knowledge sets, and can also be  connected with the business system through API  interfaces.



From the perspective of product functions, the "intelligent robot" of WeChat for Enterprise and the "knowledge question and answer" of Feishu have the same purpose, because only by connecting the private data within the enterprise and realizing calls within a safe range can the capabilities of the intelligent body be truly improved.



The difference is that Enterprise WeChat adds to the original intelligent body functions, and does not demonstrate the integration with products such as Tencent Meeting; whereas Feishu has launched a new question-and-answer function on the product, which can be accessed with one click on the product side, and deeply integrates the knowledge base, messages, and notes.



Of course, Feishu’s “Knowledge Question and Answer” is still in the trial stage, and Enterprise WeChat’s “intelligent robot” is still far from being a true intelligent entity.



2

ByteDance seizes Tencent ’s Moments




Tencent’s biggest expectation for its intelligent entity is WeChat.



At the first quarter earnings conference this year, Tencent President Martin Lau mentioned that Tencent has the opportunity to create a unique intelligent entity for WeChat, which will be deeply connected with social relationships, communications, content ecology and mini-programs within the WeChat ecosystem. "Compared with other general intelligent entities, this will be extremely unique and our core differentiated product advantage."



The outside world has great expectations for this, because the general intelligent agents of Tencent Yuanbao, ima  and other domestic manufacturers are essentially the same paths and capabilities, and it is difficult to widen the gap. As  the tasks that AI  can complete become more and more complex, only intelligent agents that combine Tencent's own advantages and capabilities can establish a leading position.



Therefore , Tencent has spared no effort in   AI  construction. Its capital expenditure in the first quarter of  2025  was 27.48  billion yuan, a year-on-year surge of  91% . It was mainly used for  AI  computing infrastructure (such as  GPU  procurement) and native  AI  product research and development.



On the other hand, ByteDance and Tencent have a consensus on the development path of intelligent entities, and are planning both general and vertical paths.



 In April  this year , Tan Dai, president of Volcano Engine, talked about ByteDance's judgment on the intelligent agent market and mentioned that if the development of AI  is to bring real changes to all walks of life, Agent  is the only way, otherwise  AI  will find it difficult to generate social and economic value.



At present,  agents  are divided into two types: vertical and general. Vertical  agents  , such as Volcano Engine's  Data Agent , focus on data analysis and intelligent marketing; general  agents  make it easier for developers and enterprises to build their own  agents . Volcano Engine has launched  the OS Agent  solution.



In Tan Dai's opinion, the definition of Agent  will become clearer this year. Similar to  the different levels of L1-L4  of autonomous driving capabilities , many  Agents  are currently only  at the L1  level, and the actual implementation may need to reach  L2++  or above.



Therefore, the outside world calls   2025  the first year of AI Agent  , essentially hoping that  Agent  will bring real changes to the industry. When the novelty of the big model itself gradually fades, more practical  Agent  will bring more frequent usage demand. Only  Agent with continuous use value can bring a commercialization path with stable traffic to cloud vendors.



Under the cloud service systems of Tencent and ByteDance, which industries have truly implemented  Agent ?



It is not difficult to see from the public agendas of the Volcano Engine Power Conference and the Tencent Cloud  AI  Industry Application Summit that Tencent focuses on smart medical care, smart education, collaborative office, and energy manufacturing, while ByteDance's industry special sessions include finance, automobiles, collaborative office, and medical health. The industries that the two are competing for include medical care, manufacturing, and collaborative office.



Especially in the automotive industry, as the largest application scenario for  AI  , it is an AI  battlefield that Tencent and ByteDance must win  .



Both parties are striving to deploy large models to automobile manufacturers and continuously expand their circle of friends. Volcano Engine has established an automobile large model ecological alliance with   25  manufacturers including FAW Hongqi, BAIC, GAC, Geely, and Great Wall. Tencent's smart travel service layout was earlier and now has more than 100 customers , including BYD, Changan Automobile, Bosch China, Yuanrong Qixing, etc.



Changan Automobile Chairman once mentioned that Changan Automobile and Tencent started their cooperation in  2018  , jointly creating a digital cockpit including WeChat, and are now jointly creating an industry-leading  AI  intelligent body. Tencent Cloud Vice President Wu Yunsheng also mentioned that customers in the automotive industry will use intelligent bodies for after-sales knowledge support and fault self-diagnosis services.



But this did not prevent Changan Automobile, which had long joined hands with Tencent, from becoming a customer of Volcano Engine.



 At the Shanghai Auto Show in April  this year , Changan Mazda launched the strategic new energy vehicle model  EZ-60 , which became the first Changan Mazda model equipped with the Doubao large model. It is said that with only voice commands, the Doubao  AI  assistant can call up more than 1,700  functions on the car  at any time. More importantly, the two parties have reached in-depth cooperation and will continue to promote the application of Doubao large models in cars.



The smart cockpit + navigation + service system  with smoother implementation is the biggest scenario for AI  application in the automotive industry. When Yuanrong Qixing, BMW, Dongfeng Motor, FAW Bestune and more and more automotive customers join the cooperation camp of Volcano Engine, Tencent Cloud can only retain old friends in the circle of friends by providing more efficient and lower-cost computing power solutions in the future.