Gemini AI Usage Billing Changes: What Google's New Limits Cost You

The Gemini AI usage billing changes went live on May 20, 2026, and they replaced a simple idea with a harder one. Instead of counting how many prompts you send, Google now counts how much computing each prompt actually costs. As a result, two people on the same plan can hit very different ceilings depending on what they ask for.

Google described the shift plainly in its own announcement: the company is moving from daily prompt limits to a compute-used model. Allowances now refresh every five hours, and those refreshes run until you reach a separate weekly cap.

Three things feed the meter: how complex your prompt is, which features you switch on, and how long the conversation has run. Because of that, a quick question and a long coding session no longer cost the same, even though both used to count as one prompt.

Google’s reasoning is defensible. A simple text prompt uses far less compute than a complex video or coding prompt, so charging both the same was always a rough approximation. Metering the actual work is fairer in principle.

In practice, though, the change is harder to plan around. You cannot see a prompt’s cost before you send it, and users report that a single heavy request can consume a few percentage points of a five-hour allowance. Instead of a number you can count, you now have a budget you can only watch drain.

Here is how the consumer tiers line up after the change. The last column is our own arithmetic: how much usage each dollar buys, expressed in multiples of the free plan.

Google also reworked the top of the range at I/O 2026. It cut the flagship Ultra plan from $250 to $200 and added a cheaper $100 Ultra tier aimed at developers and heavy creators.

Read that table again and one row stands out. AI Plus costs $7.99 for double the free allowance. In contrast, AI Pro costs $19.99 for quadruple it. Per dollar, therefore, the cheaper plan actually stretches further, and several users noticed.

However, there is a fair counterargument. Google bundles a YouTube Premium Lite subscription into AI Pro and values it at $8.99 a month. Subtract that and Pro effectively costs about $11 for four times the usage, which flips the comparison. Whether that helps depends entirely on whether you wanted YouTube Premium in the first place.

The genuinely odd row is the $200 Ultra plan. At twenty times Pro’s allowance it delivers the most usage per dollar of any tier, which is unusual: premium plans normally charge a premium.

The API side changed too, and it is stricter than the consumer side. Google now applies a monthly spend cap to every billing tier, and you climb those tiers by spending rather than by asking.

There is a second ceiling underneath those. Google also applies a spend-based rate limit on a rolling ten-minute window, set at $10 for Tier 1 and $200 for Tiers 2 and 3. Meanwhile, upgrades are quick: free to Tier 1 is effectively instant, and later steps land within ten minutes.

Google did add a meter, and it is worth bookmarking. Consumers can see their remaining allowance at gemini.google.com/usage, which shows both the five-hour window and the weekly cap.

Developers get a separate view. Once a billing account is linked to a project, the AI Studio billing page tracks spend against the tier cap. Neither view predicts what a prompt will cost, so the practical habit is to check before starting anything long rather than after.

Much of the frustration is about framing rather than mathematics. Google’s announcement email compared the Pro plan to the free tier instead of to what Pro subscribers had the week before, so the message read as an upgrade to people who had just been cut back.

The timing did not help either. Newer models such as Gemini 3.5 Flash consume more resources per request, so the meter empties faster on exactly the models Google is promoting. For anyone who used Gemini for long coding or research sessions, the practical experience is a tighter ceiling arriving without warning. If that describes you, it is worth comparing the best AI writing tools before renewing.

If metered pricing has you rethinking the whole arrangement, running a model on your own hardware removes the meter entirely. See how a Google Gemma AI mini PC handles it for around $300. And if you would rather build something more capable, start with the best GPU for AI.

*Sources: Google, Google AI for Developers, 9to5Google, Engadget, best-ai.news.*