Google introduced compute-based Gemini usage limits at I/O 2026 just last week. The idea was to track how much processing power each request uses rather than counting individual prompts. It sounded reasonable on paper. Then users actually tried using it. The backlash was quick. Subscribers reported hitting their five-hour caps after just a handful of prompts. In some cases, after just one. One Google AI Pro subscriber posted video proof showing a single avatar video generation attempt draining their entire five-hour allowance before the video even finished. Gemini lead Josh Woodward responded publicly and said he’d look into it. Google has now followed up with a round of fixes specifically in response to feedback about the Gemini usage limits being too tight. Here’s What’s Changing The biggest fix targets Gemini 3.1 Pro. Google is capping how much quota a single prompt can use, so one heavy request can’t wipe out a full session’s budget. Failed requests also won’t count against your limits going forward.…