Google has revamped its AI usage measurement for Gemini applications, shifting from fixed prompt counts to a system that limits users based on the computational resources their queries consume. This change, effective since May 17, results in many users reaching their limits faster than before.
Under the updated model, rather than counting the number of questions posed to Gemini, Google tracks the processing demand each query requires. For example, simple queries like checking the weather now consume far less quota compared to complex tasks such as analyzing lengthy documents. Users who engage with the advanced Gemini 3.1 Pro model, designed for intensive workloads, report hitting usage caps significantly sooner.
Details of the Compute-Based Quota System
The new limits reset on a rolling five-hour basis but also include a cumulative weekly cap to prevent users from circumventing restrictions through cooldown periods. Factors influencing quota consumption encompass the complexity of prompts, the use of additional tools during sessions, and the overall length of interactions.
Shortly after these measures were introduced at Google I/O 2026, the company responded to user concerns by limiting the maximum impact any single prompt can have on weekly allowances and excluding failed requests from quota deductions around May 28. They also introduced lighter Gemini 3.1 Flash-Lite models for certain scenarios at no cost and hinted at future pay-as-you-go credit options for AI services.
This update clearly differentiates between user tiers: free accounts maintain baseline limits, Google AI Plus subscribers receive double the quota, Pro subscribers get four times the baseline, and Ultra tier subscribers are allowed up to twenty times the standard allowance. The wide range shows a significant divide between free and premium users.
Although Google has not integrated on-chain billing or blockchain-based metering for Gemini, this move supports a usage pricing approach already employed by some crypto-native AI projects. The adjustment may influence pricing strategies across the AI and decentralized computing sectors.
