Reporter |
Editor | Wen Shuqi
Douba is about to start charging, and many people see this as a turning point.
“After the land grab, it’s time to monetize, or the costs will be unbearable.” A large model developer said. Another investor told the reporter that after the AI products have gone through the competition of DAU, they will inevitably transition to competing on business value.
On May 4th, Douba’s App Store page announced that it will launch a paid version with more value-added services and revealed a three-tier subscription price of 68 yuan, 200 yuan, and 500 yuan per month. If calculated annually, it would be 688 yuan, 2048 yuan, and 5088 yuan.
According to the reporter’s understanding, the specific pricing and official launch time have not been finalized. Douba officials also responded that they will continue to provide free services and explore more value-added services on top of the free services. The relevant plan details are still in the testing phase.
Currently, Douba’s user base has become the absolute first in the domestic market, far surpassing Qianwen, DeepSeek, and Yuanbao. According to Questmobile data, in March 2026, Douba’s monthly active users reached 345 million, Qianwen had 166 million, DeepSeek had 127 million, and Yuanbao had 57.35 million.
However, the massive user base also corresponds to higher reasoning costs. In April, Douba’s daily token calls exceeded 12 billion. Unlike the internet field, where more users mean thinner cost distribution, in the AI field, more users mean greater power consumption (electricity, hardware depreciation) costs.
Using Douba’s Seed 2.0 lite version (there are also pro and mini versions) input 0.6 yuan/100 million tokens, output 3.6 yuan/100 million tokens API pricing, if all commercialized, this would be a daily revenue of over 1 billion yuan, with costs of at least tens of millions of yuan.
On the other hand, it’s not just the cost of power consumption that’s affecting, but the rise of complex task demands and the gradual formation of high-value user groups are more fundamental factors.
This year, OpenClaw’s popularity worldwide and the subsequent “shrimp farming craze” in China proved that under suitable harness engineering, large models can already perform more complex, diverse, and automated task processing capabilities. At the same time, various manufacturers’ text, voice, image, and video multimodal models have also shown significant performance improvements.
For model manufacturers, they can identify the proportion of people with high demands for complex tasks through the data in their hands. These users have more professional usage scenarios, higher task quality requirements, and a certain tolerance for cost expenditure, making them the professional users that all enterprises in the AI commercial ecosystem are looking for.
“These paying users have a clear account in their minds.” An entrepreneur with AI community operation experience told the reporter, especially for demands related to creation and research, there is usually a clear cost reference, “As long as the efficiency improvement is significant compared to traditional processes, they are willing to pay.”
In this case, not distinguishing between light and complex tasks for users leads to frequent lag, delays, or hallucination problems during use, which affects the experience and evaluation of the model for this part of the users — in the long run, this may have a greater impact on Douba’s commercialization performance.
“Complex scenarios consume more resources.” An insider close to Douba told the reporter, “For example, the PPT generation scenario, because many people use it, and the resource consumption is high, it takes a long time to queue.”
Some model manufacturers have already reacted to this. Kimi adjusted its membership rights system this year, and one of the changes was to switch from “pay-per-use” to “pay-per-actual-consumption”. When paying per use, complex tasks and simple tasks consume the same amount, but after the adjustment, they are charged according to the actual situation of task difficulty and consumption.This reflects Kimi’s awareness of the increasing proportion of complex task consumption.
Therefore, the launch of the subscription version can help Douba recover some of the power consumption costs and serve as a screening to identify professional users who are willing to pay for efficiency. Referring to Anthropic and OpenAI’s soaring ARR, this is also a natural move to gain revenue after the free user acquisition cycle.
However, many doubts still surround Douba.
The biggest question is whether Douba, which has always positioned itself as a life assistant, can also support the label of AI productivity.
Overseas, Anthropic and OpenAI’s revenue has been rising, essentially due to their flagship models and products like Claude Code and Code X, which have established a market position in AI productivity. After two or three years of market education, plus a good payment environment, they have achieved commercial success.
In the domestic market, startups like Zhishu, Kimi, and Minimax have faced greater revenue pressure, so they have set their product targets at developers, creators, and other groups with a higher willingness to pay. Douba, Qianwen, and Yuanbao, on the other hand, have been pursuing user breadth, and the AI battle during the Spring Festival this year was essentially a competition for a national-level AI entrance.
The risk of Douba charging at this point is that if its model capabilities do not create an absolute gap with free competitors, it may lead to the loss of high-value users in the short term.
The reporter learned from an informed source that Douba’s subsequent versions will have significant improvements and changes in productivity and Agent scenarios.
OpenAI’s transformation may provide some reference for Douba. At the beginning of this year, OpenAI predicted that its profit structure would undergo a significant change in 2026, as the ChatGPT Go, priced at only $8, would lead to a surge in paid subscription users to 122 million, while the ChatGPT Plus subscription users would decline by 80%, and the professional version Pro subscription users would double.
At the same time, the company’s advertising revenue from its massive user base would surpass subscription services, predicting that by 2030, it would generate approximately $10.2 billion in revenue, becoming the largest source of income, accounting for about 36% of the total revenue.
This may not be the path Douba wants to take, but it does provide a reference for balancing scale and professionalism.
As a large factory with a massive app ecosystem, ByteDance’s Douba can have more imagination space in subscription services. For example, if it evolves into a super member, it can link with professional rights and interests on platforms like Jianying, Feishu, and Douyin, and its scale advantage and scenario advantage can still be played out.
In late January, at ByteDance’s first all-hands meeting, CEO Liang Rubo explicitly stated that “Douba/Dola” AI assistant is the company’s current core goal, a peak that must be climbed in the short term.
This highlights the significance of Douba’s adoption of a paid model. After all, it’s not just technology and scale, but commercialization is also a crucial step it must take.
