On May 4, a number of media reported that the Doubao App quietly hung up the subscription instructions for the paid version on the page of the Apple App Store.

In response to this situation, Doubao officially responded that these payment plans are still in the testing stage, and the specific details will be released through official channels when they are officially launched. At the same time, the official also emphasized that the familiar free services will continue to be retained and will not be canceled.

|Paid function focuses on productivity scenarios

At present, when you open the Doubao App, you can't find any paid options or feature entrances. Doubao officials also explained that these paid plans are only in the testing stage, and when they are officially launched, complete information will be released through official channels.

据知情人士透露，未来的付费功能主要是为了解决一些复杂的工作任务，比如自动生成PPT、做深度的专业数据分析，以及制作影视内容等。

To put it simply, as AI becomes smarter and smarter, it can help everyone handle many difficult tasks, but these complex tasks will consume more computing power and reasoning time, and the cost is relatively high, which is the main reason for the launch of paid services. As for our daily use such as chatting and checking information, the free version will always be retained, and you can continue to use it with confidence.

|Token usage is a thousandfold, and the cost account behind the bean bag payment

The fundamental reason behind Doubao's attempt to charge this time is actually the huge computing power cost brought about by the rapid growth of AI large models.

The data can intuitively reflect this growth: According to Volcano Engine, by March this year, the number of tokens (which can be simply understood as units of information processed by AI) processed by the Doubao model has exceeded 120 trillion yuan per day. This number has doubled in the past three months and has skyrocketed 1,000 times compared to when it was first released in May 2024. At the same time, on the Volcano Engine platform, the cumulative usage of more than one trillion enterprise customers has also increased from 100 at the end of last year to 140.

Tan Dai, head of ByteDance's Volcano Engine, previously explained that the difference in token prices is essentially due to their different carrying capacity. He pointed out that the next generation of AI models will be smarter, and although the cost of a single token will increase, the economic value it can create will also increase simultaneously.

To put it simply, the "price increase" brought about by the improvement of model intelligence is essentially because it can help customers solve more difficult problems and create greater value.

This actually explains why Doubao charges - when the AI model's capabilities are strong enough to help everyone complete high-value professional tasks (such as doing complex PPT or data analysis), it is reasonable to charge a certain fee to users with professional needs.

This can not only meet the needs of professional users, but also alleviate the huge pressure on computing power costs, so as to better balance the contradiction between mass popularity and cost control.

|Doubao follow-up payment, and the commercialization layout of domestic large models has been basically completed

From a global perspective, paid subscriptions for AI products have long become mainstream.

For example, foreign ChatGPT is divided into multiple levels such as free, Plus, and enterprise version, and well-known large models such as Anthropic and Grok have already opened paid subscriptions. In China, leading manufacturers such as Zhipu and DeepSeek have also launched paid versions before. Therefore, Doubao's follow-up charging this time means that the paid layout of domestic mainstream large models has been basically completed.

However, it is worth noting that domestic large models do not have a "unified pace" in their pricing strategies, and everyone has their own way of playing:

On the one hand, Zhipu has chosen to increase prices continuously this year. They raised API prices three times in succession in February, March and April, with overall increases ranging from 10% to 30%. After the price adjustment, the price of some of its products has approached the level of the well-known foreign model Claude.

On the other hand, DeepSeek went in the exact opposite direction, starting a "big price cut". At the end of April, DeepSeek not only opened a limited-time 2.5% discount on some models, but also announced that the cache hit price of the full range of API services was directly reduced to one-tenth of the original. The latest price is as low as about 0.02 yuan per million tokens.

It can be seen that whether it is a price increase or a price reduction, it is an independent choice made by each manufacturer according to its own market positioning, and there is no unified trend in the industry at present.

According to the analysis, Doubao's exploration of payment this time is mainly to find a balance between its huge user scale and commercial monetization, which is the independent choice of the enterprise, and does not mean that the entire industry is collectively turning to price increases.

Jingtai View|AI commercialization enters the "deep water area"

Doubao quietly hung up the "price list", which is not only a business choice for an enterprise, but also a microcosm of the development of the domestic AI large model industry. When the user scale reaches 100 million and the number of calls reaches trillions, it is difficult to maintain a long-term policy by relying solely on free policies.

As the model's capabilities improve, the pricing of tokens will be more tied to the economic value it creates. AI applications that can solve complex tasks and create high-value AI applications will have stronger pricing power and user willingness to pay.

Domestic large model manufacturers have shown a differentiated trend of coexistence of "price increase" and "price reduction" in their pricing strategies. This reflects the differences in technical strength, cost control and market positioning of various manufacturers, and the industry competition will be more fierce and diversified.

Pay attention to AI manufacturers with strong technical strength and cost control capabilities

In the process of commercial transformation, AI manufacturers that can continue to improve model efficiency, reduce computing power costs, and launch high-value payment functions will be more likely to stand out in the fierce market competition and achieve sustainable commercial monetization.

Pay attention to the upstream and downstream related enterprises of the AI industry chain

With the acceleration of the commercialization process of AI large models, the demand for upstream and downstream related enterprises in the industrial chain such as computing power infrastructure, data services, and AI application development tools will also continue to grow. These companies are expected to benefit from the overall development of the AI industry.