Google overhauls Gemini usage limits with compute-based system

Google is changing how it measures your weekly Gemini AI usage, moving away from simple daily request limits to a more complex “compute-based” system. The shift reflects how advanced AI features are breaking traditional flat-rate subscription models across the industry.

The new system considers multiple factors when calculating your usage. Instead of counting basic requests, Google now factors in prompt complexity, which features you use, and how long your conversations run. This means generating images or videos, using deep research tools, or accessing advanced thinking models will eat into your allowance faster than simple text queries.

Google announced the changes in a support document, though the company remains vague about exactly how the new calculations work. What’s clear is that paid users get significantly higher limits than free users, with the differences being substantial.

The new tier structure shows how much Google values its paying customers:

• Google AI Plus ($8/month): 2x standard usage limits
• Google AI Pro ($20/month): 4x standard usage limits
• Google AI Ultra ($250/month): 20x standard usage limits

Your usage limits will refresh every five hours until you hit a weekly cap. This represents a major change from the previous system, where Google AI Pro users simply got 100 Gemini Pro requests per day, regardless of complexity.

This move puts Google in line with broader industry trends. GitHub made similar changes to its Copilot plans less than a month ago, switching from “premium request units” to “AI Credits” based on actual token usage. The pattern reflects a common challenge across the AI industry.

The shift comes as AI companies struggle to manage the computational demands of advanced “agentic” features. These tools can spawn multiple sub-processes that consume tens of thousands of tokens from a single user request. Traditional flat-rate pricing simply doesn’t account for this variable resource consumption.

Not everyone is tightening the screws, however. Anthropic recently doubled the limits for its Claude Pro and Max plans, but only after securing additional computing capacity through a deal with SpaceX. An Anthropic executive admitted last month that their current subscription plans “weren’t built” for resource-intensive features like Claude Code and desktop AI agents.

This trend toward usage-based pricing signals a maturing AI market where companies are getting more realistic about the true costs of advanced AI features. For users, it means more complex pricing but potentially fairer allocation of computational resources based on actual usage patterns.