The Black Market for Tokens in China
The Black Market for Tokens in China
A researcher from Oxford has uncovered how Chinese providers trade tokens at prices lower than official rates, employing various tricks.
A researcher from the Oxford Chinese Policy Lab wrote an article about how Chinese providers trade tokens at prices lower than official ones. Sometimes, the price for access to a model can be as low as 10% of the official rate. The question arises: how?
It turns out there are three options. Choose your favorite ☕️:
Mass Data Collection. Such providers gather all requests, responses, and reasoning chains – and resell them. That's how they profit.
Model Substitution. You are sold Opus 4.7, but in reality, they substitute it with some locally produced Qwen. Thus, the discount on a powerful model turns into profit from a cheap one.
Subscription Arbitrage. For example, they take Claude Max for $200 and split it among dozens of clients due to token limits per hour. Again, profit! This also includes generating numerous fake API accounts to obtain free starter credits, fake student accounts, and reselling unused quotas from others.
All geo-blocking, phone verifications, foreign card requirements, and everything else are easily bypassed by these proxies. Thus, the market has grown, and official authorities can do nothing about it.
Chinese users massively take advantage of this, as direct access without VPNs and foreign cards is closed to them. And everyone seems satisfied.
Why it matters
AnalysisThis black market for tokens highlights serious access issues to technology in China, as well as challenges for regulators trying to control the situation.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
Vibe Coding for Non-Developers: How Marketers, Designers, and PMs Build Products with Claude and Cursor
A practical guide to Vibe Coding without the terminal: 45 techniques for working in Claude and Cursor tailored for marketers, designers, and PMs. From website edits to routine automation.
Hermes Profile: Growth Marketer & Acquisition Funnels
Hermes growth marketing lead profile: product positioning, AIDA/PAS copywriting frameworks, conversion funnels, competitor analysis, and launch campaigns.
Groq LPU Processors (500 Tokens Per Second)
The Groq Language Processing Unit (LPU) is a new type of specialized computing processor that generates responses from large models at speeds of 300–500 words per second by eliminating slow external memory (HBM) and utilizing ultra-fast SRAM. This architecture accelerates text generation to unprecedented levels.
Output Tokens Limit (Why Text Cuts Off Mid-Word)
A hardware or software limit on the maximum length of a single response (Max Output Tokens). This explains why long code or articles sometimes stop mid-sentence and how the magic phrase 'Continue' brings the model back to work.