Gemini stops metering by number of prompts and switches to compute: a window refreshes every 5 hours, with a weekly cap layered on top. Note that in agent mode or the Gemini CLI, a single prompt can fire dozens of model requests internally.
Original text +
Gemini 不再按提示词条数计量，改成按算力计：每 5 小时刷新一个窗口，外面再套一层每周上限。注意在 agent 模式或 Gemini CLI 里，一次提示可能在内部触发几十个模型请求。
A change to the limit rules themselves, affecting your allowance going forward.