How Gemini's new usage limits work and where to check them

A Gemini conversation no longer works like a simple message counter. On May 17, 2026, Google changed how it presents and enforces Gemini Apps limits: they are now defined as compute-based limits. That means two seemingly identical requests can consume different shares of the quota depending on the model, the chosen feature, the complexity of the prompt, and the length of the conversation.

For daily users, the consequence is plain: there is no fixed "one message, one credit" equivalence. A short request to the fast model does not carry the same capacity cost as a deep research run, a long conversation with files, or a task that uses extended reasoning. Google groups all of that under the same notion of compute usage.

How the limit refreshes

Google's official help says the limit refreshes every five hours until a weekly cap is reached. Running out therefore does not always mean waiting a full week: capacity may return at the start of the next window. But the five-hour reset does not remove the weekly maximum, and actual availability can shift with demand.

Google also warns that it may change limits without notice due to capacity constraints. In periods of high activity, some demanding features may become unavailable first for accounts without an AI plan. The explanation avoids false precision: the numbers a person sees in their own account, and the moment a notice appears, matter more than any table copied from the internet.

Plans change the relative level of access. The official documentation places Google AI Plus at twice the standard limit, Google AI Pro at four times the standard, and Google AI Ultra at five or twenty times the Pro limit, depending on the subscription. That comparison describes relative access, not a promise of an identical number of responses for every feature.

And one fact is worth underlining precisely because it is absent: the official page publishes no absolute figures. It does not say how many messages the standard tier allows, how many deep research reports fit in a day, or what "twice as much" equals in concrete units. The multipliers refer to a standard that Google does not quantify in public. That opacity is information in itself: any table of exact message counts circulating on forums or social networks does not come from the official documentation, and can go stale the moment Google adjusts capacity.

A limit for every feature

Gemini Apps is not the only product that consumes capacity. Google explains that each product has its own limits. Using Gemini in the chat, creating content in Google Flow, working with an agent in Antigravity, requesting deep research, or using Gemini inside Gmail and Docs does not necessarily draw from a single pool of messages.

The clearest proof that each product keeps its own currency is on Google's subscriptions page: Flow, the audiovisual creation tool, is measured in points — 200 a month on the Plus plan, 1,000 on Pro, and between 10,000 and 25,000 on Ultra — while NotebookLM promises Pro users "five times more" audio overviews than the free tier. Files have their own ceilings too: the same page cites uploads of up to 1,500 pages per document or 30,000 lines of code. None of those figures is deducted from the same counter as a chat conversation, and that is exactly why it pays to check each feature's quota rather than a single one.

Plans also change the context size available in Gemini Apps: 32,000 tokens without a plan, 128,000 with AI Plus, and one million with AI Pro or AI Ultra. A larger context lets you hand over more material at once, but it does not guarantee that a long task costs the same as a brief query. In fact, the documentation itself lists conversation length as one of the limit's factors.

When a feature's cap is reached, Google may offer to continue the conversation with Flash-Lite, to wait until the quota refreshes, or to move to a plan with higher limits. Pro and Ultra subscribers can also buy AI credits to extend usage in Flow, Antigravity, and other compatible products. Those credits do not turn every feature into unlimited consumption: they work where Google accepts them and add to a base quota tied to the plan. And the drop to Flash-Lite is a quality trade-off, not an arbitrary punishment: it is the lightest step in the family, with faster, cheaper-to-serve answers in exchange for less reasoning capacity.

What each tier costs and what it includes

As of this revision, July 31, 2026, the subscriptions page for Spain lists four tiers: free, Google AI Plus at 4.99 euros a month, Google AI Pro at 21.99, and Google AI Ultra from 99.99, with a higher tier at 219.99. Model access scales too: without a plan, the page offers Gemini 3.6 Flash and "variable access" to the Pro model; Plus and Pro give access to the Pro model with expanded limits; Ultra adds the highest level of access and early entry to features the page lists as Deep Think and Gemini Spark. Deep research, by contrast, appears listed even on the free tier, though subject to the general compute limits.

Comparing prices also requires looking at what is not artificial intelligence. The plans bundle storage — 400 gigabytes on Plus, five terabytes on Pro, twenty or more on Ultra — and packaged subscriptions such as YouTube Premium, in its Lite variant for Plus and Pro. Part of the monthly fee pays for things that have nothing to do with Gemini's limits, and it is worth mentally discounting them before deciding whether the jump in tier is worth it for the AI alone.

Where to find the useful information

The most reliable way to check your personal status is to open Gemini on the web, go to Settings, and select "Usage limits." Google says the app notifies you when you approach your limit and again when you reach it, including when it will refresh. That screen is the operational reference because it reflects your plan, your region, and the capacity of the moment.

Three questions are worth separating before signing up for a plan: which model and features you need; how much context length you require; and which product actually generates your consumption. Someone who only asks brief questions may get no value from a higher tier. Someone who works with long documents, research, or creative tools should review both the Gemini Apps limits and the quotas of those specific features. With the prices in view, the math is less abstract: if what you need is context — moving from 32,000 tokens to one million — the relevant jump is to Pro; if what you need is volume in Flow or video generation, the points per tier matter more than the general multiplier.

Google's change does not make the quotas easier to summarize, but it does make them more transparent about what determines them. The unit is no longer the isolated message: it is the amount of resources a specific interaction requires. Understanding that difference — and knowing that the only reliable figure is the one on your own limits screen — helps explain why the same account can get more or fewer responses on different days and tasks.