Google $GOOGL announced several changes to Gemini's usage limits after subscribers complained about hitting caps with only a handful of prompts.
The company switched to a compute-based usage system at I/O 2026 last week, moving away from the previous prompt-based limits. Rather than counting individual prompts, the system weighs factors like request complexity, tool usage, and conversation length to determine how quickly a user's quota is consumed.
Among the changes taking effect now, Google said it is placing a ceiling on the amount of quota any single prompt can consume when using Gemini 3.1 Pro, a response to users finding that complex requests with large files rapidly drained their allowances. According to Gemini lead Josh Woodward $WWD, the fix addresses reports of complex requests draining allowances rapidly: the company is now "capping the amount of quota a single prompt can use so you get more out of the Pro model."
Interactions using Flash-Lite have been carved out of the quota system altogether, so those exchanges no longer draw down a user's allowance. The company also clarified that failed requests will not be charged against a user's quota.
A bug that had been eating through Ultra subscribers' quotas after generating only a small number of Omni videos has been resolved, and as a result those users now have access to twice as many Omni video generations.
Because tasks like Deep Research draw heavily on compute resources, Google plans to surface richer tracking information and notifications so subscribers can better manage their budgets; today the gemini.google.com/usage page shows only a broad overview.
On model persistence, Google clarified that a chosen model stays locked in by default; the only circumstances under which the app will switch to a different one are a manual override by the user or an automatic fallback triggered by hitting a usage cap.
Looking ahead, Google said it plans to let Gemini users purchase pay-as-you-go top-up AI credits, though no timeline was provided for that feature.