Google imposes compute‑based caps on Gemini; users report throttling

Google has started enforcing a new kind of limit inside Gemini: not a simple message count, but a compute meter. Since May 17, the company has told users that Gemini Apps usage now depends on the complexity of a prompt, the models and features used, and the length of a chat, according to a support page updated for personal Google accounts. Users on May 21 began posting complaints that the new system was harder to predict than the old one, with some saying they were being throttled sooner than they expected. The change lands days after Google I/O 2026, where the company introduced new Gemini features and adjusted its paid AI plans. At the same time, Google has tied some of those features more tightly to subscription tiers inside Google One. That has turned a product question — what Gemini can do — into a packaging question as well: which plan unlocks which tool, and how often a user can actually use it. ### How do the new Gemini limits work? Google’s support documentation says Gemini Apps now use “compute-based usage limits” rather than simpler usage caps. The company says those limits are affected by three variables: prompt complexity, the model or feature selected, and the length of the conversation. The same page says a user’s allowance refreshes every five hours until a weekly limit is reached. The help page does not give consumers a single universal number of prompts per day. Instead, Google says users can upgrade to a Google AI plan for expanded access to models and features in Gemini Apps. That wording leaves the practical limit dependent on behavior rather than a fixed quota. ### Why are users saying they are getting throttled? 9to5Google reported on May 21 that users were already frustrated after the new limits went live, saying the change made consumption feel more metered and less transparent than before. (support.google.com) The outlet said limits had not been a major concern for many Gemini users before I/O 2026, but became more visible once Google shifted to the compute-based system. A separate 9to5Google report later on May 21 said Google had already raised limits for Antigravity, its agent-focused development platform, after user concerns. The publication said Google tripled those limits twice, suggesting the company was still tuning the thresholds after rollout. ### Which features now depend on which Google One plan? Android Authority reported on May 21 that Google’s Google One lineup became more complicated after I/O 2026, with some new features, including Gemini Spark, unavailable on lower AI tiers. (9to5google.com) The publication said Google also introduced a new $100 AI Ultra tier between AI Pro and the previous $250 Ultra plan, while cutting the old top-tier price to $200. (9to5google.com) Google’s own help pages say Google AI Ultra members get first access to new models, prioritized traffic and the highest usage limits for Antigravity. A separate Gemini help page says paid Google AI plans offer expanded access to Gemini features and models. ### What did Google announce around Gemini at I/O? Google said at I/O 2026 that it was entering what Chief Executive Sundar Pichai called the “agentic Gemini era.” In keynote and product posts, the company highlighted Gemini Omni, Gemini 3.5 Flash, updates to the Gemini app and new agent tools including Antigravity. (androidauthority.com) Those announcements expanded what Gemini can do, but they also widened the gap between headline capability and actual access. (support.google.com) Google’s public materials now describe a system where model access, feature access, traffic priority and usage ceilings vary by plan. ### What should users watch next? May 17 is the date Google lists for the Gemini limit changes, and May 21 is when early consumer complaints became visible in tech press coverage. (blog.google) The next concrete signal will likely come from Google’s own support pages and plan descriptions, where the company has been updating language on limits, upgrades and tier benefits. Google’s current documentation points users to Gemini Apps Help and Google One Help for the latest plan-specific details, including refresh timing, weekly caps and which features sit inside AI Pro or AI Ultra. (support.google.com)