DouBao launches paid subscription plans, redefining AI pricing standards

Article by Next Trend, Author | Fang Yuan, Editor | Xiao Yu

In May 2024, Tan Dai, President of Volcengine, announced at a conference that the pricing for the Doubao large model was significantly lower than industry rates, at just 0.0008 RMB per thousand tokens, eliciting gasps from the audience.

Two years later, Doubao unveiled a new paid subscription plan, introducing three paid tiers, with the highest tier priced at 500 yuan for continuous monthly billing.

The most aggressive player in the price war has率先 launched a new fee structure.

And the competitors' first response was not to follow suit, but to emphasize that their services were still free.

On May 4, 2026, Doudou added subscription options on its latest App Store page, introducing three paid tiers, with the Standard plan priced at 68 yuan and the highest tier reaching 500 yuan.

The message immediately rose to number one on Weibo's hot search list.

Some users in the comments said they would uninstall if there are fees, others claimed DouBao is exploiting retail investors, and some are concerned that the free version will be severely degraded.

Two years ago, Tan announced at the Volcano Engine FORCE Power Conference that the input price for the Doubao general model Pro-32K was 0.0008 yuan per thousand tokens, just one one-hundred-fiftieth of the industry average.

At that time, the audience in the crowd did the math: at this price, one yuan could buy 1.25 million tokens’ worth of input, equivalent to three copies of "Romance of the Three Kingdoms."

This pricing has completely overturned the industry.

Alibaba Cloud has reduced the input price for Qwen-Long, its flagship model, from CNY 0.02 to CNY 0.0005, a decrease of 97%.

Baidu, on the other hand, has taken a more straightforward approach, making its two flagship models, Speed and Lite, completely free.

A price war ensued, and its impact quickly spread throughout the industry.

The survival space for startups has been severely squeezed—either dive deeper into vertical industries or build up an application ecosystem; the generic middle layer no longer exists.

Shen Jue, President of Baidu Intelligent Cloud, once said: "The malicious price wars in China's large model industry have caused overall revenue to be orders of magnitude lower than that abroad."

The industry has undergone significant changes today.

According to calculations by Guolian People's Securities, China's overall daily token consumption surged from the tens of billions level at the beginning of 2024 to the quadrillions level in February 2026.

Amid surging demand, Zhipu AI, Tencent Cloud, and others have successively issued price increase notices, with some products seeing hikes of over 400%.

The underlying logic of a price war is to burn cash for market share, but when the market size reaches its peak, burning cash is no longer an investment—it’s a waste.

Tan Dai responded on his social media circle regarding the large model price war, emphasizing that the pre-training cost of Doubao 1.5 Pro is significantly lower than that of other domestic models, and that it still maintains substantial gross margins at current pricing. He had previously stressed in an interview that Volcano Engine never sacrifices profits to gain market share.

If true, then over the past two years, DouBao hasn't been selling at a loss to gain attention—it has been dominating competitors with its cost advantage.

According to QuestMobile's Q1 2026 report, Doudou's monthly active users reached 345 million, while Qianwen, in second place, had 166 million, a difference of over 180 million.

Meanwhile, Doubao added 101 million new users in a single quarter, accounting for 80% of the industry's total new users.

Over two years, Doubao's monthly active users grew from 26 million to 345 million, increasing 13-fold and far surpassing all competitors.

When DouBao became virtually synonymous with the entire category, its pricing effectively set new industry standards.

However, it is worth noting that the 345 million users are not all direct paying customers.

Baidu's goal is not to have everyone pay, but rather to support its growth needs with a small percentage of users opting to pay, given its large user base.

In addition to its large user base, DouBao also demonstrates impressive user retention.

Relevant data shows that DouBao’s 30-day retention rate is 44.5%, significantly higher than Qwen’s 23.5% and YuanBao’s 30.1%; the average monthly usage frequency for DouBao is 54.8 times, three times that of Qwen’s 19.8 times; in terms of activity rate, DouBao also leads with 33.5%, compared to Qwen’s 17.1%.

Each one represents a gap of overwhelming magnitude.

High retention rates indicate that users face very high switching costs when confronted with fees; although users may claim they’ll uninstall if charged, once they do uninstall, what alternatives are there?

As of March 2026, DouBao's daily average token usage has exceeded 120 trillion, a 1,000-fold increase since May 2024.

But a 1000-fold increase in usage comes with a 1000-fold increase in computing costs.

ByteDance's capital expenditure in 2025 will exceed RMB 150 billion, with approximately RMB 90 billion allocated to AI computing power; in 2026, it plans to invest RMB 160 billion, of which RMB 85 billion will be used for AI chip procurement.

High expenses led to a year-over-year net profit decline of more than 70% in 2025.

Li Liang, Vice President of ByteDance, clarified that, when calculated based on operating profit margin, the decline was actually minor; however, he acknowledged that the slowing growth of TikTok Commerce, coupled with increased investment in new businesses, has indeed created pressure.

Doubao's free model is not unsustainable; rather, each additional day of operation erodes the profit margin of its core business.

In the first quarter of 2026, the global monthly active users of AI applications surpassed 2.7 billion, with China accounting for over 50% of the new growth, bringing the domestic monthly active user base for AI applications to 851 million.

Essentially, users who currently meet the conditions for using AI have already become users of AI applications.

Once the guest has received everything, free becomes a cost rather than an investment.

After DouBao announced its pricing, competitors responded more aggressively.

DeepSeek was the first to announce a limited-time 25% discount on its V4-Pro model API; Alibaba’s Qwen launched a subsidy campaign offering users free milk tea to attract attention; Yuanbao and Wenxin Yiyan directly promoted free usage.

Doubao's pricing table has put all competitors in a difficult position: charging fees is tantamount to handing users over to rivals, but maintaining free service means admitting they are still in the user education phase.

The price of Doubao Standard Edition at 68 yuan per month was not set arbitrarily.

According to Tencent Research Institute’s 2025 survey data, approximately three-quarters of AI users have already paid or are willing to pay, and among those who have paid, more than 55% spend less than 100 yuan per month. A price point of 68 yuan precisely aligns with the psychological affordability range of most users.

Has there been a precedent for this pricing? Yes.

Kimi has validated the entire industry. Within less than a month of launching the K2.5 large model, its cumulative revenue exceeded the full-year total for 2025, with global paying users increasing fourfold month-over-month. Stripe’s data is even more astonishing: the number of orders from Kimi’s individual subscribers surged 8,280% month-over-month.

The valuation of Moonshot AI surged to $10–12 billion, setting the fastest record in China for a company to reach unicorn status since its founding.

Chinese AI users aren't unwilling to pay—they just didn't have products worthy of that price before.

With 345 million monthly active users, even if only 1% convert, the scale will far exceed Kimi's entire paid user base.

An earlier precedent dates back ten years. In 2016, WeChat Pay became the first to charge a 0.1% fee on cash withdrawals. At the time, Alipay still led the mobile payment market, and many believed that WeChat Pay, without having established a decisive advantage, was giving up its competitive edge by introducing fees so early.

No users abandoned WeChat Pay as a result. The core use cases of WeChat Pay, such as sending red packets, splitting bills, and transferring funds for repayment, have strong user retention and can only be completed within its ecosystem.

The user's exit cost is extremely high—so high that fees cannot possibly serve as a threat.

Today, DouBao is replaying the same scene: the platform with the largest user base charges first, setting not its own profit margin, but the price benchmark for the entire category.

Morgan Stanley has already done the math. Based on a paid conversion rate of 0.3% to 3% and monthly active users ranging from 345 million to 525 million, DouBao's annualized subscription revenue is estimated to be between $100 million and $1.5 billion, with a base-case scenario yielding revenue of approximately $426 million to $684 million.

However, Wall Street’s optimism hinges on one assumption: an annual renewal rate of 30% for domestic C-end tool products is already considered top-tier in the industry, while similar overseas products typically achieve renewal rates above 60%. Whether DouBao can convert first-month impulse purchases into long-term retention remains unanswered.

At the same time, OpenAI across the ocean took a different path.

ChatGPT has 500 million weekly active users, with only 25 million paying users, resulting in a paid conversion rate of approximately 5%, meaning only 1 out of every 20 users is willing to pay.

OpenAI's solution is to launch a low-cost version of ChatGPT Go, priced at $8 per month, while integrating advertisements and setting a 2026 advertising revenue target of $2.5 billion. By using advertising revenue to subsidize free users, OpenAI aims to minimize the barrier to paid access.

Doubao chose the opposite path: rather than relying on ads that dilute the user experience, it directly charges power users for advanced features.

Behind the divergence of the two paths lies the assessment that a purely free model is unsustainable. One chooses advertising as a fallback, while the other opts for subscription services to support operations.

Morgan Stanley provided a clearer qualitative assessment: the phase of cultivating Chinese consumers' AI usage habits has largely been completed, and the industry is shifting from user subsidies to commercially sustainable growth.

If DouBao’s paid model proves successful in the market, competitors like Tongyi and Yuanbao, which continue to offer free services, will face a difficult choice: follow suit with pricing or stick to their free strategy.

Two years ago, Doubao disrupted the industry’s original pricing model with a rate of 0.0008 yuan per thousand tokens; two years later, it rebuilt this model with a price range of 68 to 500 yuan.

The participant who once broke the rules is now the leading force in reestablishing them.

The first person to publish a price list did not define their own profit margin, but established a clear price benchmark for the entire industry.