The way people use artificial intelligence is changing. Instead of counting how many prompts you submit, Google Gemini is now focusing on how much computing power your requests consume.
Google has introduced a new Usage Limits system for Gemini, allowing users to track their usage based on AI compute resources rather than the number of prompts they send. This marks a significant shift in how AI services are managed.
Previously, the system was straightforward: one prompt generally counted as one use. Asking for today’s weather and generating thousands of lines of code had little difference in how usage limits were calculated.
That is no longer the case. For example, a simple request like “What’s the weather today?” requires very little computing power. On the other hand, generating complex code, analyzing large documents, using Deep Think features, or creating AI-generated videos requires significantly more processing resources. As a result, these tasks consume more of a user’s available usage quota.
The reason behind this change is the growing computational cost of modern AI models. Features such as million-token context windows, advanced coding assistance, image generation, and video creation rely on massive GPU clusters and large-scale data centers. Running these capabilities for millions of users requires enormous computing resources.
Instead of measuring how many prompts users send, Google is now measuring the actual workload placed on its infrastructure. Gemini now includes a dedicated Usage Limits page where users can monitor two key metrics:
- A rolling 5-hour compute limit
- A weekly usage limit
If the 5-hour limit is reached, users must wait until it resets before accessing certain advanced features again. If the weekly limit is exhausted, some subscription plans may temporarily switch users to a lighter AI model until the quota is restored.
Google also assigns different compute limits across its subscription tiers. Higher-tier plans receive access to more computing resources, although the company does not disclose the exact number of prompts or compute units included in each plan. Since every request requires a different amount of processing power, usage is measured dynamically rather than by a fixed prompt count.
Users can track their remaining quota directly from the Usage Limits dashboard, where both short-term and weekly usage are displayed in real time. Once a limit is reached, it automatically resets after the specified period, or users can upgrade to a higher subscription tier to access additional compute capacity.
The goal of the new system is to manage AI usage based on the complexity of each request and the actual computing resources required, rather than simply counting the number of prompts submitted.
