Chen Bai's article The news that tofu buns will be charged has caused a thousand waves with just one stone.

On May 4, 2026, Doubao updated its service statement on the product page of the Apple App Store, disclosing its payment plan for the first time. According to the page information, Doubao will launch three payment levels - Standard Edition, Enhanced Edition, and Professional Edition, corresponding to different pricing.

Although the payment model is currently only a value-added service of Doubao, and its free basic services will still be retained, this is still interpreted by the outside world as: domestic large models have bid farewell to the era of "traffic for users" and officially entered the deep waters of commercial closed loop.

If the previous "Hundred Model Wars" focused on parameters, computing power, and even gimmicks, then from the moment when the Doubao Plan charges, the competitive dimension has quietly returned to the most primitive and cruel business principle: value exchange. The problem that has been tormenting OpenAI all along is now facing domestic AI (artificial intelligence): how can AI companies make money?

Previously, most domestic large models maintained an unspoken understanding: no matter how high the backend cost was, the frontend was always free, attempting to exchange economies of scale for capital favor and user data. And the payment plan of Doubao has finally broken this unsustainable balance. The reason is not complicated: the consumption of tokens is simply too expensive.

In the era of AI, for large model companies, every time a user clicks "send", there is a real consumption of computing power behind it. This consumption is irreversible and extremely expensive. More frankly, the larger the number of users, the higher the cost - even for Internet giants, this may be unbearable for a long time.

In fact, looking at the world, AI giants in Silicon Valley have already taken the lead in the matter of "paying". Whether it's OpenAI's ChatGPT Plus, Anthropic's Claude, or Google's Gemini, a clear subscription model has been established almost from its inception or early iteration stages.

The logic of Silicon Valley is very simple; The computing power and research and development costs of large models are almost astronomical. If they cannot achieve self generation through a healthy business model, the so-called "AGI dream" will eventually become a money burning game. After more than a year of free long-distance running, domestic big models have also realized that if they do not bow down to reality, they cannot keep up with the future.

However, at the beginning of this new era, the most thorny question is facing everyone: who would be willing to pay for the paid model of domestic AI?

From the perspective of business history, the domestic Internet market really lacks the tradition of paying for software for a long time. In the past twenty years, we have become accustomed to exchanging watching advertisements for free videos. In the subconscious of most people, software and services should exist like air, naturally for free.

When AI upgrades from a "chatbot assistant" to a "productivity tool that requires monthly payment," users' psychological games will be extremely intense. This kind of game not only exists in the choice of whether or not to pay, but also in the comparison between domestic AI and the world's top level.

This touches upon the core threshold of charging - alignment of abilities. It can be foreseen that once entering the payment stage, users' tolerance will rapidly decrease. When the product is free, users may be able to laugh off occasional "hallucinations" or mediocre answers; But when they need to pay tens or even hundreds of yuan per month, their demand will no longer be "being able to speak", but "having a conversation".

For Doubao and all subsequent domestically produced AI that follow up on charging, what they need to deliver is no longer just an answer, but a solution that satisfies users and has certainty.

At present, the alignment of capabilities is a crucial step that domestic large models must take towards charging. The user's choice logic is extremely simple: if they are inferior to Claude 3.5 in logical reasoning, inferior to GPT-4o in creative writing, and lagging behind Gemini in code writing, why should they give this money to domestic models?

The implementation of the charging model is essentially pushing domestic AI directly onto the international stage. Paid users are the most demanding reviewers: they will vote with their feet to choose the "cheat brain" that can truly improve efficiency and solve practical pain points.

Therefore, the fee for tofu buns is more like submitting a "nomination letter". It promises users that since they have received payment, it will provide stronger computing power support, better algorithm strategies, and deeper scene adaptation. This is undoubtedly a good thing for improving the overall quality of domestic AI, as it forces companies to shift from "emotional value" to "effectiveness", and from focusing on "user scale" to "retention rate" and "average customer value".

Will you pay for the bean buns that start charging? This is not just a question for users, but also a test question for all domestic AI practitioners. Before the answer is revealed, the commercialization of domestic AI has just begun.

(The author is a senior media professional)