The Token Transit Station is a unique product of the AI era, serving as an intermediary layer between users and large models by aggregating access to nearly 20 domestic and international large models such as ChatGPT and Claude, helping users bypass regional restrictions and access AI services at prices lower than official rates.

Author: Zhan Fangge

Source: Baobian

Reselling "digital currencies" in the AI era: Token intermediaries race forward—but destined to be short-lived?

When you ask an AI to write a weekly report, the underlying large model receives your request, responds to your message, and understands the context—all of which are broken down into individual units called tokens, translated into Chinese as “词元”.

Each interaction with AI is measured and priced in tokens. As the "hard currency" of the AI era, the flow of tokens is reshaping narratives around wealth distribution.

Some people don’t create tokens—they simply act as intermediaries, moving tokens around and profiting from the price spread, making substantial profits. This business is known as a token relay station.

Some companies achieving monthly transaction volumes in the tens of millions may have teams of fewer than 20 people. “While traditional industries are still figuring out how to turn a profit and cut costs, these companies spend their days thinking about how to avoid taxes,” said a relatively new Token intermediary station operator describing his peers’ business.

Initially, token relay stations primarily existed to circumvent regional restrictions on overseas large models, using various methods to connect these models within China for local users. However, as this business has evolved, many relay stations are now also "exporting" effective domestic large models to serve overseas markets.

Behind the token intermediary station lies not only tales of wealth creation, but also a story of demand outpacing regulation.

01 Business on the Trend

How do you understand a token intermediary?

For example, to watch certain variety shows or movies, you might need to subscribe to memberships on different platforms like Tencent, iQiyi, Youku, and Bilibili. If you subscribe to all of them, your actual usage rate may be low, and it’s a waste of money—this is when you’d naturally wish for a shared video membership platform where, for the price of one membership, you could access all these resources freely.

The token intermediary specializes in aggregating large models, and it employs various methods to efficiently obtain tokens from different large models at low cost, allowing users to access domestic and even region-restricted overseas models like ChatGPT, Claude, and Gemini for less money.

“If your token usage isn’t very high, the cost of using a relay station will be lower than subscribing directly to an account—sometimes significantly so,” a relay station operator told Bao Bian. Some users employ relay stations not only to bypass regional restrictions imposed by large model providers but also because this method is more affordable and allows them to experiment with a wider variety of large models.

On the token relay interface, up to nearly 20 large models can be invoked; each model has different billing methods and corresponding rates. Some are so cost-effective that millions of tokens can be purchased for just a few yuan, whereas subscribing to ChatGPT Plus costs $20 per month.

The low technical barrier and constant gray-market tactics upstream have led more and more people to try entering this space for a share of the profits, driving down token prices.

The 21st Century Business Herald previously reported that the official API output price for Claude-Opus-4-6 is approximately RMB 170 per million tokens, with some domestic intermediaries offering prices at half the official rate, and smaller platforms reducing it further to 20-30% of the official price.

In fact, the token intermediary business dates back to 2024 or earlier. “In 2023, people were already reverse-engineering the ChatGPT 3.5 model. The release of Claude 3.5 marked a qualitative leap in vibe coding—natural language programming—and since then, programmers have had massive demand for tokens,” AI knowledge monetization blogger McDonald told Baobian.

AI technology is increasingly permeating daily life and work, transforming industries across the board, while demand for tokens continues to surge.

Previously, CCTV reported that in early 2025, China's daily average token usage was only 7 trillion, rising to 30 trillion by mid-year, and surpassing 100 trillion by the end of 2025. In early 2026, the daily average token usage reached 140 trillion.

This is only the usage volume according to official figures; when including demand for overseas model calls, Token consumption will be even higher. Thus, in today’s world where even college students rewriting their theses want to use top-tier models, Token intermediaries have sprung up like mushrooms after rain.

Surprisingly, the technical barrier for the token relay station itself is not high. Ma indicates that vibe coding has significantly lowered the barrier to setting up a relay station—station operators only need basic technical knowledge to deploy their sites.

“There are two clear open-source options available: one called sub2 API and another called New API,” Ma Dian said. “You just need to buy a server and tell it, ‘This project looks good—deploy it on this server for me,’ and the AI will handle it for you.”

Taking the New API as an example, it is a comprehensive solution integrating protocol conversion, channel management, billing, and user backend management, available for users to call at will.

The technical barrier is low, but the business of being a token intermediary is both appealing and risky.

02 Gray Chain

To become a token intermediary, you need to focus on the upstream supply chain—specifically, how to access overseas large models at a lower cost.

The common approach is the wholesale-retail token model. Major platforms typically offer developer subscription plans (Coding Plans), which usually include a set number of free API calls, with additional usage priced in tiers. Intermediaries can bulk-purchase multiple Coding Plans, consolidate the quotas into a shared account pool, and resell them at a price slightly lower than the official API rates.

For example, the official API sells at 1 yuan per 10,000 tokens, while the Coding Plan reduces the cost to only 0.2 yuan per 10,000 tokens; the intermediary resells at 0.5 yuan, yielding a gross profit margin of 60%.

This approach has strong compliance, but profits depend on volume.

The station manager mentioned above told "Bao Bian" that around 2024, overseas large model providers were still in the customer acquisition phase, offering free usage quotas to small AI startup companies with the goal of fostering long-term collaboration at the application layer.

“The station manager said this may represent the initial form of domestic token relay stations and one of the lowest-cost methods for accessing overseas large models.”

Today, the business model of token intermediaries has given rise to numerous gray-market practices, the most typical of which is "reverse trading."

For example, software like Windsurf and Cursor have built-in access to large models such as Claude. Some intermediaries reverse-engineer these models and connect them to their own relay platforms for user access. “It’s like stealing water and electricity—hooking up your own pipe to a public water line to use utilities from legitimate sources for free,” said the operator of one such relay platform.

Additionally, due to individual subscription plans offered by companies such as GPT and Claude, some intermediaries first acquire large numbers of individual subscription accounts and then, through reverse engineering, split the calling capacity of these accounts to sell to more users.

During this process, obtaining batch number sources is likely linked to gray-market activities such as credit card fraud and Apple ID theft. Additionally, traffic diversion may lead to account throttling and a degraded user experience.

For users, while token intermediaries may seem convenient and cost-effective, the risks involved can be substantial. For instance, users often fund their accounts upfront and are insensitive to token consumption, allowing intermediaries to secretly deduct fees higher than agreed upon. However, Xiao Zhang, an industry professional in AI, told "Bao Bian" that users can easily spot such practices if they look closely—so only intermediaries unconcerned about their reputation would engage in this behavior.

In addition, “挂着羊头卖狗肉” is also very common—“replacing a high-quality model with a cheap one is called watered-down,” Ma said. The greater risk comes from data. Ma explained that (reverse proxying) involves using technical means to forward users’ questions to a large model, then intercepting the model’s response and sending it back to the user.

Throughout the process, user data is not only fully under the control of the intermediary but also circulates on the website, creating opportunities for hackers.

Xiao Zhang noted that most relay stations will label whether a service is "reverse" after calling a large model. However, even if labeled as "claude-officially," whether it truly is an official API service provided by the large model vendor largely depends on the站长's integrity.

The 21st Century Business Herald previously reported that many transit station operators were asked whether they could sell user numbers, and it was also revealed that several major companies in the industry acquire data to train their models.

As an AI professional, Xiao Zhang is aware of these risks. He only uses the intermediary station for a small portion of his work, such as modifying code or documents, and never discloses sensitive data or passwords. The rest of his work is completed using the domestic large model provided by his company.

For the vast majority of users in other industries, being aware of data risks themselves may already be a barrier.

03 Industry Differentiation

On one hand, this long-standing business operating in a gray area has been labeled as high-risk; in May 2026, a domestic AI relay station operator was criminally detained for 37 days by public security authorities for illegally reverse-engineering and reselling low-cost AI interface resources.

On the other hand, in the same month of May, the sequential entry of two high-profile figures brought Token intermediaries into the spotlight.

On May 1, prominent crypto figure Sun Yuchen announced on social media the launch of his "strongest AI intermediary ever"—B.AI, emphasizing blockchain login, purely anonymous payments, zero tampering, and the lowest prices online. However, B.AI requires payment in cryptocurrency and deliberately bypasses identity verification processes, drawing significant regulatory scrutiny.

On the same day, Fu Sheng, Chairman and CEO of Cheetah Mobile, officially launched EasyRouter. Shortly afterward, the company issued a statement on its official website stating that it does not provide services within China and supports refunds. Fu Sheng also changed his description of EasyRouter from “relay station” to “enterprise-grade model gateway/routing layer.”

The official website states that its API sources are directly procured from cloud service providers and AI platforms. Leveraging top-tier cloud providers and AI platforms, it delivers compliant, stable, and auditable AI computing power distribution services.

In other words, EasyRouter states that its upstream suppliers are fully compliant and that it has not resold the models to users in other regions against the wishes of overseas large model companies.

Currently, upstream large model providers are facing increasingly stringent regulations. On June 12, 2026, local time, Anthropic announced that it had received instructions from the U.S. government, compelling the company to impose comprehensive access restrictions on its two most advanced AI models, Fable 5 and Mythos 5, covering all institutions and individuals outside the United States.

As compliance becomes increasingly stringent, domestic companies like EasyRouter are also beginning to explore compliance in this business.

In fact, a compliant Token business is not complicated. In 2023, OpenRouter was established in the U.S. as an aggregation platform for multiple large model providers, leveraging their substantial Token usage volume to secure better pricing from the platform and then reselling to users. In simple terms, it’s wholesale-to-retail Token trading.

Beyond acting as a wholesaler, there are fully B2B business opportunities with lower compliance risks. The station manager mentioned above told "Bao Bian" that he does not primarily profit from C2C station operations, but rather from helping others set up stations. For example, many business owners are willing to pay for the deployment of a non-profit station within their company so that employees can access better AI models.

Some are also attempting to provide peripheral services for transit stations, such as cryptocurrency standardized billing and smart reconciliation services for fund management. A professional in this field told "Bao Bian" that while domestic small- and medium-sized token transit stations rarely require cryptocurrency settlement services, many large overseas token transit stations do have such needs. According to industry standards, the commission rate is approximately 1%.

It also noted that these large token intermediaries do not merely facilitate the integration of overseas large models into China; they also handle the export of domestic large models. “DeepSeek, Doubao, and some models from Alibaba’s ecosystem have a significant advantage in being cost-effective—by not selling to the U.S. but targeting other markets, they hold a decisive advantage.”

Thus, the Token intermediary has become an honest slice in the AI flood, faithfully recording how real demand flows and shifts.

"The code itself has little value in this era," said Xiao Zhang.

Similarly, what truly determines how far the token transit business can go isn’t hidden in the code. During the industry’s wild growth phase, whoever controlled the supply held the advantage—but as upstream manufacturers crack down more strictly on violations, the profit margins built on reverse-subscribed channels and black-market account sources are destined to shrink ever further.

Ultimately, the emergence of token intermediaries was likely just to fill a market gap during this unique period—a temporary product of the AI wave. If one day the technological barriers around AI are dismantled, that may be when intermediaries begin to fade away.