Developments in AI technology have now resulted in AI agents being able to replace multiple people performing multiple tasks simultaneously. The increased attention caused the AI agents to start consuming five times more tokens than humans. Considering the pricing policies of Chinese AI service providers, this may give them a market advantage.
Image source: Unsplash, Igor Omilaev
Artificial intelligence agents work independently and perform a series of complex operations autonomously. They access various services and databases and then check the results of their work. All these operations require the use of tokens – traditional data processing units. go through According to OpenRouter aggregatorproxy queries generated 15 times more tokens than standard human-generated queries. Furthermore, by the beginning of August, token consumption on the platform had increased 14 times compared to what it had been six months earlier.
Agent activity exceeded human activity for the first time in February; by August, the number was more than five times higher. Proxy queries tend to have higher data duplication if they are retrieved from cached information, but such operations require less computing resources, so providers charge less for them. However, load growth rates are so high that the material costs of maintaining ecosystems grow rapidly. Taihe Capital experts said that the use of tokens by artificial intelligence models will grow exponentially in the future.
If by the beginning of 2024, the number of tokens used in China every day is 100 billion, then by June this year, this number will increase five thousand times to 500 trillion. In this case, customers will try to allocate less important workloads to lower-cost models because the cost of outsourcing all the work to higher-end systems is too high. In this sense, China’s AI models have lower user costs and are likely to increase their customer base and revenue.
In comparison, China’s DeepSeek V4-Flash values 1 million standard input tokens at $0.14 and 1 million output tokens at $0.28. OpenAI US offers GPT-5.6 Sol at $4 per million input tokens and $20 per million output tokens. Tokens for Anthropic Fable 5.1 cost more: $10 and $50 respectively. Statistics from the OpenRouter platform show that the DeepSeek solution is the most popular, with a request share of 22.7%, while the US Google and OpenAI are 20.4% and 19.6% respectively.
If you find an error, select it with your mouse and press CTRL+ENTER.