<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/353be265-e79f-4205-af63-22ff8c2d2016.jpeg?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/353be265-e79f-4205-af63-22ff8c2d2016.jpeg"/> Author | Wang Xiaojun
Editor | Huang Yu
The competition in China's AI field is still reflected in the actions of internet giants. As the year-end approaches, major companies are intensively summarizing their achievements and releasing new moves.
&#34;The daily token usage of the Doubao large model has exceeded 50 trillion, an increase of more than 10 times compared to the same period last year.&#34;
On December 18, at the Volcano Engine FORCE Power Conference, the data announced by Volcano Engine President Tan Dai attracted industry attention. As of December this year, over 100 enterprise clients have accumulated token usage exceeding one trillion.
This means that ByteDance has officially joined the top competition in the global token economy. On the same day, ByteDance also released the Doubao large model version 1.8 and the audio-video creation model Seedance 1.5 pro, fully upgrading its multimodal Agent capabilities.
To further enhance its competitiveness and attract talent globally, ByteDance announced on December 19 a salary increase for all global employees, aiming to attract top professionals in the AI field.
Of course, other players in the industry are also accelerating their efforts.
A few days before the Volcano Engine FORCE Power Conference, the industry was quite lively. Tencent officially announced organizational restructuring and brought in former OpenAI researcher Yao Shunyu as Chief AI Scientist; the &#34;AI talent&#34; and post-95 technical expert Luo Fuli, who previously joined Xiaomi, made her debut, introducing the specifics of Xiaomi's large model; Alibaba, while heavily promoting its C-end applications Qianwen and Lingguang, recently launched Afu targeting the health sector...
From the actions of various companies, it can be seen that the AI battlefield is rapidly evolving from early technical parameters and model scoring to a multi-dimensional war concerning top talent, industrial ecology, and commercialization capabilities.
Each giant has chosen its own breakthrough path based on its endowments, but without exception, they are all heavily betting resources on the competition for future intelligent ecological dominance. The outcome of this competition no longer depends on breakthroughs in single-point technology, but on who can first complete the construction of an ecosystem from model capabilities to industrial empowerment, from technical talent to commercial closure.
<h2>01 ByteDance's Aggressive Push</h2>
At this conference, the Doubao large model version 1.8 was officially released. The new version has significantly enhanced its tool invocation capabilities, complex instruction adherence, and OS Agent capabilities, upgrading the model from an &#34;efficient executor&#34; to a &#34;versatile planner.&#34;
In terms of visual understanding, the frame count for single video understanding in Doubao 1.8 has doubled from 640 frames to 1280 frames, supporting low-frame-rate understanding of ultra-long videos and enabling high-frame-rate understanding of key segments through tool invocation. This capability allows AI to accurately locate accident scenes and analyze the vehicles and times involved, which can be widely applied in safety inspections, product quality inspections, and other fields.
However, in light of the current updates to the Doubao large model, Volcano Engine President Tan Dai believes that there is still a certain gap between the current Doubao large model and Gemini 3. However, he also stated that the effectiveness of catching up depends on the measurement dimensions
&#34;From a distance perspective, we do have gaps; from a short-term speed perspective, there are also shortcomings, but our acceleration is improving, and we will gradually close the gap in the end.&#34;
In addition to model updates, in response to content creation needs, Volcano Engine has launched the Seedance 1.5 pro audio and video creation model, achieving a technological breakthrough. This model adopts an innovative native audio and video joint generation architecture, achieving millisecond-level audio-visual synchronization output.
Seedance 1.5 pro supports various elements such as ambient sound, background music, and human voices, supports multi-person multi-language dialogue, and has precise lip-sync alignment, covering Chinese dialects, English, and minority languages. This technical detail is of great significance for globalized creation, which also shows ByteDance's deep layout in the content creation field.
It is worth mentioning that after the recent emergence of the Doubao phone, it has sparked a major debate in the industry regarding privacy and security. At the beginning of this conference, Volcano Engine started by discussing security.
Tan Dai believes that security is the most fundamental condition for using AI. Furthermore, in the field of AI security, Volcano Engine has officially launched a new feature: MaaS on AICC. Through this service, all models on the Volcano Ark, including the Doubao general model family and open-source models like deepseek, can run directly in the AICC encrypted environment.
With this feature, users only need to select the &#34;confidential deployment&#34; method in Volcano Ark to easily enable confidential inference services for various models with one click.
Tan Dai previously emphasized that the price reduction of Volcano Engine is not about losing money for market share, but rather achieving cost optimization through algorithms, software and hardware, and engineering methods.
In the past two years, Volcano Engine has completed a curve overtaking. Before 2025, Volcano Engine was not considered a mainstream cloud market vendor. However, in the past year, Volcano Engine has attracted many B-end and C-end users by gradually lowering the price of the Doubao general model.
For the industry, Volcano Engine can provide industry-leading models on one hand, and on the other hand, it can offer high-performance, low-cost AI infrastructure. With the support of both aspects, the token consumption of Volcano Engine has rapidly increased. In this year's institutional reports, it has already become the number one vendor in the public cloud market share.
To further enhance competitiveness, on December 19, ByteDance announced a salary increase plan. This includes four major measures: increasing bonuses (including performance options) investment, raising the performance evaluation cycle for the entire year of 2025 by 35% compared to the previous cycle; significantly increasing salary adjustment investment, raising it by 1.5 times compared to the previous cycle; and raising the lower limit (starting salary) and upper limit (ceiling) of total compensation for all job levels.
This round of salary increase is not unrelated to the AI competition in the industry. As competition intensifies, leading players are accelerating the recruitment of AI talent. Previously, there were market reports that some companies had offered double salaries to poach ByteDance's AI researchers. Now that ByteDance has officially announced the salary increase, it is evident that it is also aimed at retaining and attracting more AI talent.
<h2>02 Industry Race</h2>
Summarizing the development of Volcano Engine over the past few years and becoming an important player in the current AI competition, Tan Dai candidly stated that the core reason is catching up with the technological windfall of AI
In his view, &#34;It is crucial for technology companies to seize opportunities; when the opportunity has not yet arrived, the work done is mostly incremental, with growth rates possibly only at 20%-30%. However, when the opportunity comes, significant changes can be seen within one or two years. In fact, there is not much difference between the two; it is just the timing that leads to different effects.&#34;
However, the current opportunity belongs to ByteDance and the industry. Recently, leading players in the industry have been rolling out their latest chips to position themselves for a new round of competition.
On December 17, Tencent upgraded its large model research and development architecture, newly establishing the AI Infra Department, AI Data Department, and Data Computing Platform Department to comprehensively strengthen its large model R&amp;D system and core capabilities.
At the same time, Tencent officially announced that Yao Shunyu (Vinces Yao) would serve as the Chief AI Scientist in the &#34;CEO/President's Office,&#34; reporting directly to Tencent President Liu Chiping; he will also concurrently serve as the head of the AI Infra Department and the Large Language Model Department, reporting to the President of the Technology Engineering Business Group, Lu Shan.
On the same day, at the 2025 Xiaomi &#34;Human-Vehicle-Home Full Ecology&#34; Partner Conference, Luo Fuli, head of the Xiaomi MiMo large model, known in the industry as an &#34;AI prodigy,&#34; made her &#34;debut&#34; on stage, officially releasing the open-source Xiaomi MiMo-V2-Flash large model.
Another player, Alibaba, has been particularly active in AI this year. Since November, it has fully entered the AI to C market. Official data shows that Qianwen surpassed 30 million monthly active users (including APP, Web, and PC) within 23 days of public testing, becoming the fastest-growing AI application globally. Recently, Ant Group announced that its AI health application AQ brand has been upgraded to &#34;Ant Afu,&#34; starting AI transformations in various vertical fields.
In the face of industry competition, Tan Dai has shown a relatively optimistic attitude.
He stated, &#34;Next year, the most important thing between models is not competition; it is about expanding the market. This market is expected to grow tenfold next year. What everyone is looking at is not the competition for existing shares, not a zero-sum game, but rather how we can grow this market together.&#34;
Moreover, he believes that lowering prices is also very important. &#34;Volcano was one of the first in the country to emphasize the importance of models; we need to look at tokens. They were the first to say that model prices should drop significantly through technology, reducing them by 99%, and we still have a gross profit. Others didn't believe it, thinking we must be losing money, but now that everyone has lowered prices, they believe it.&#34;
However, competition in the industry has always transitioned from blue ocean to red ocean, encroaching on each other's territories. The controversy sparked by the emergence of Doubao mobile phone is an example of this.
Tan Dai believes that the so-called &#34;conflict&#34; between APP and Agent is not obvious; the key lies in perspective. &#34;From the user's point of view, the core demand is to conveniently and cost-effectively meet their own needs. The value of AI lies in amplifying overall demand. For example, if someone previously found it troublesome to go downstairs to buy coffee and reduced their purchases, now, through AI, they can conveniently place orders and have them delivered, possibly increasing their consumption from one cup every two days to two cups a day, significantly increasing demand.&#34;
It is evident that AI competition is shifting from breakthroughs in technical points to the construction of ecological systems. Behind the daily usage of 50 trillion tokens for the Doubao large model is ByteDance's comprehensive layout in computing power, data, and application scenarios
In the qualifying matches, everyone is putting in their utmost effort. With the saturation of investment in the industry, future competition will inevitably become more intense. In the future, competition in the industry will continue to unfold around how to transform AI technology into actual productivity and how to build a sustainable AI ecosystem

BYTED

First, raise salaries for employees

Doubao surges, ByteDance's AI strikes again