As time goes on, the models drain much faster. You used to be able to prompt without worrying about limits, especially for small tasks. Now, a single prompt takes up 15% of the 5-hour limit before it even generates an output?
Agents are incredibly expensive, Google first provided higher quotas to attract market, but currently they are trying to turn some profit
If you are on a pro plan, it might be helpful to try out Open Code instead. It doesn’t have the same quality, but you can run the agents for much longer. Antigravity is stronger in my opinion
It totally depends on your plan and usage. Google AI Pro subscription isn’t suited for heavy daily workday agentic development, but is well-suited for experimenting with it. If you are on an Ultra plan, then you’re doing something wrong because it is hard to melt your quota to zero with it (even harder on the 20x one). If the 15% decrease is due to using a Gemini model, then I assume you used Gemini 3.5 Flash or Gemini 3.1 Pro at a high reasoning level, or a third-party model (sounds like Claude Sonnet; Opus usually consumes 33% of your weekly quota). You usually only want to use high reasoning levels while planning, not as the main coder/s. Related to third-party models: model quota multipliers only affect Google models, not third-party ones.
Regards