Increase of up to 400%! Tencent Cloud officially announces AI model price adjustment "Lobster" goes viral, potentially intensifying Token inflation

Today, Tencent Cloud Intelligent Body Development Platform announced that it will optimize and adjust the billing strategy for some models.

According to the announcement, this adjustment mainly involves two types of changes. One is that the public beta models will no longer be free; models such as GLM 5, MiniMax 2.5, and Kimi 2.5 will end their free public testing on March 13 and transition to official commercial services, with billing based on model usage. The other change is an increase in prices for the Tencent HY2.0 Instruct and Tencent HY2.0 Think services from the Mixyuan series. The specific details are shown in the chart below:

[Image source: Tencent Cloud]

As seen, both input and output prices have generally increased by more than four times. For example, the input price for Tencent HY2.0 Instruct was originally 0.0008 yuan per thousand tokens and has been adjusted to 0.004505 yuan per thousand tokens, a rise of 463.13%.

Looking across the industry, several major model providers have signaled price increases. Recently, OpenAI raised the input price for GPT-5.4 to $2.50 per million tokens and the output price to $15 per million tokens. Zhipu AI increased the GLM Coding Plan subscription price by up to 60%, with API call prices rising by 67% to 100%.

On February 15, the Daily Science and Technology Innovation Board reported in “AI Meets the Strongest Spring Festival Blockbuster: Token Inflation Has Become Inevitable?” that demand for tokens is experiencing “inflation.” According to Guolian Minsheng Securities, this “inflation” does not simply mean that tokens are becoming more expensive; it also refers to a structural increase in token consumption per unit time and per user.

Recently, the open-source AI “Lobster” OpenClaw became a global sensation shortly after launch. As a personal AI agent assistant that can run on users’ own devices, its token consumption increased more than fourfold in just one month. Data shows that China’s overall daily token consumption is expected to exceed 30 trillion by mid-2025, and by February 2026, the combined daily token consumption of mainstream large models has reached hundreds of trillions.

Guosheng Securities pointed out that due to the natural multi-tool invocation, long context, and multi-process work characteristics of agents, the token consumption of AI agents is quite high. As the scale of intelligent bodies and task complexity increase simultaneously, active AI agents’ token consumption is entering a rapid growth phase. According to IDC forecasts, the number of active intelligent agents in China will surpass 350 million by 2031, with token consumption experiencing an exponential increase of over 30 times annually.

The aforementioned institutions further emphasized that with the sharp increase in token consumption, many mainstream model providers are facing computational power shortages, further increasing demand. The explosive growth of OpenClaw’s penetration verifies the commercial viability, and the rising penetration rate and complexity of agents drive a surge in token consumption, creating rigid demand for computing power. As token consumption in the domestic market explodes, domestically produced computing power is expected to gradually dominate at the infrastructure level due to cost advantages and a complete ecosystem.

(Source: Cailian Press)