Industry

Gemini usage limits explained: what 2x, 4x and 20x actually get you

Google sells Gemini plans as 2x, 4x and 20x. Here is what those numbers are measured against, and what actually burns through your limit.

Multi Chats Team
September 10, 2026 · 6 min read

You open Google's plans page. AI Plus says 2x. AI Pro says 4x. AI Ultra says up to 20x.

Twenty times what, exactly?

The page will not tell you. So here it is, in the order you actually need it.

Google stopped counting messages in May

On 20 May 2026, straight after Google I/O, the Gemini app moved to compute-based usage limits.

Your allowance stopped being a message count. It became a budget.

Google's help centre says what spends it. The limits "factor in the complexity of your prompt, the models and features you use, and the length of your chat". Your limit then "refreshes every 5 hours until you reach your weekly limit".

The practical version: a run of short text questions barely registers. One long video generation can eat your afternoon.

9to5Google reported the switch on 21 May 2026. Google's help page now says the same.

The multipliers, in plain English

Two Google pages publish these numbers. They do not agree.

The help centre lists the tiers plainly. Without an AI plan you get "standard limits". AI Plus is "2x higher than standard limits". AI Pro is "4x higher than standard limits". AI Ultra is "5x or 20x higher than AI Pro limits depending on your subscription".

The plans page tells it differently. Ultra is sold there as "up to 20x", under a footnote reading "compared to people without a Google AI subscription".

Those are not close. If Ultra is 20x Pro, and Pro is 4x free, Ultra is eighty times the free tier, not twenty. Both pages were live when we checked them on 1 September 2026. Google has not reconciled them publicly.

Plus and Pro are the settled part. Both pages agree that 2x and 4x are measured against the free tier.

Treat Ultra's number as unknown until Google says which of its own pages is right.

The bit that annoyed people

Here is the trap in that structure.

Pro's 4x is measured against free. The comparison Pro subscribers actually wanted, against what Pro gave them before 20 May, appears nowhere.

So a subscriber who was already paying saw a big number and no way to tell whether their own allowance had gone up, stayed flat or gone down.

Users raised exactly that, plus a second objection: per dollar, AI Plus buys more Gemini usage than AI Pro does. Both were reported by 9to5Google on 21 May 2026.

Google has never published an absolute number. Only multiples of a baseline it also declines to quantify.

What Google fixed after the backlash

To Google's credit, it moved fast. On 28 May 2026 it announced a set of adjustments.

  • A single prompt can no longer drain your whole quota. There is a cap on what one request can consume.

  • Failed requests stopped counting. Google's wording: "If a request fails, you won't be charged. Our system mistakes are on us, not you."

  • Prompts to 3.1 Flash-Lite became free against your limit. Google's help page now tells subscribers who run out that they can carry on in Flash-Lite.

  • More detailed usage breakdowns were promised.

  • Video generations for Ultra were doubled.

  • Your model choice now persists between sessions.

  • Pay-as-you-go AI credits were confirmed as coming.

That is a real improvement. It still does not give you a number.

Gemini Notebook joins the same system on 2 September

This is the fresh bit. Gemini Notebook moves onto compute-based limits on 2 September 2026, announced 28 August. As of this writing on 1 September, it has not switched over yet.

What changes for Notebook users:

  • Daily limits are gone. Limits now reset every five hours, same as the main app.

  • A usage cost indicator bar appears under Studio generations, showing what an output is expected to consume before you run it.

  • If your selection would exceed what you have left, Notebook suggests a lighter output instead.

  • Heavy jobs like Video Overviews and Slide Decks get a "Generate later" option. They queue and run automatically once your limit resets.

  • You will see notices like "You're almost at your AI usage limit. Limit resets at 3:00 PM" and "Limit reached. All features are available after 3:00 PM."

The tier structure carries straight over, Ultra's ambiguity included.

How to work out which tier you actually need

Google will not do this maths for you, so do it backwards.

Ask what you generate, not how often you chat. Text questions are cheap. Video, slide decks, long document analysis and image work are not.

The five-hour window matters more than the month. Most people who hit a wall hit it in a burst. If you work in long sessions, the reset window beats the multiplier every time.

Try Plus before Pro. Plus is 2x free. If 2x is enough, 4x is money you did not need to spend.

Ultra only pays off if you generate video. The doubled video allowance from 28 May 2026 is the clearest concrete benefit Google has attached to that tier.

The new Notebook cost bar is the best teacher available. It is the first place Google has shown you what an individual action costs. Watch it for a week and you will learn which of your habits are expensive.

One European footnote

Check what your money buys where you live. Google's plans page, read on 1 September 2026, marks Gemini Agent, one of Ultra's headline features, as US and English only. The AI Mode hotel booking tools launched on 27 August 2026 were US English only too. The tier is global. Its contents are not.

The pattern behind all of this

A multiplier is a comparison, and a comparison needs two numbers. Google publishes one of them, and it publishes that one twice, differently.

Anthropic sells the same shape, Max 5x and Max 20x, measured against its own Pro plan rather than a free one. So the habit is not Google's alone. Google is simply the case where two of the company's own live pages contradict each other.

Google's help centre adds one line worth reading before you pay: "Limits may change without notice, including due to capacity constraints."

Where that leaves you

Track your own usage rather than the label, upgrade one step at a time, and re-read the plans page after any Google announcement. That number has already moved twice in four months.

Long chats are one of the three things Gemini charges you for, so a thread running for weeks costs more per message than a fresh one. Our explainer on what a context window is covers why.

If you would rather not have your usage tied to one company's undisclosed baseline, that is roughly why MultiChats exists. Sixty models from eighteen providers sit behind one subscription. The free plan does not expire. It runs on 50 messages every 24 hours, two image generations and two web searches per 30 days. Plans are in the pricing section.

Figures in this piece are current as of 1 September 2026. Google has changed this system twice in four months, so check the live plans page before you buy.