Gemini Usage Limits in 2026: How the 5-Hour Refresh and Weekly Limit Really Work

Short answer: as of November 2026, Google does not publish a fixed prompts-per-day number for the Gemini app. Its Gemini Apps Help page on usage limits describes limits that depend on how much compute your requests use, that refresh every 5 hours, and that stop at a weekly limit. Free users get "standard limits," Google AI Plus gets 2x, AI Pro gets 4x, and AI Ultra gets 5x or 20x of Pro.

That is the whole official picture, and it is vaguer than most blog posts make it sound. Google does not say how many prompts "standard" is, and it does not state the weekly limit as a number. If you saw a figure like "5 prompts a day" or "100 prompts a day" somewhere, it did not come from the page that governs the app today.

The quick version of what Google says

• Limits are compute-based. They depend on prompt complexity, the models and features you use, and how long your conversation is.
• The limit refreshes every 5 hours "until you reach your weekly limit."
• Paid plans raise the ceiling by a multiplier, not by a published number of messages.
• Google can change the limits without notice, including because of capacity constraints.

A note on how this post is built

Every claim about Gemini's behavior below comes from two Google pages: the Gemini Apps Help article and the Gemini subscriptions page. Where Google is silent, I say so instead of filling the gap with a number from a blog. Facts are dated "as of November 2026," and the last section shows you how to re-check them, because these pages change.

If you have read about Gemini limits before, you probably saw simple tables: a handful of "thinking" prompts a day on the free tier, a larger number on a paid plan, and a bigger number again on the top tier. Those tables were built when Google published per-feature daily caps. The current help page does not contain them.

What changed in how Google describes limits

The current help page frames usage as a budget that different requests spend at different rates. A short text question costs very little. A long running conversation or a large file analysis costs a medium amount. Deep Research and video generation cost the most. The page says video "generally requires the most usage."

When the cost of a request varies that much, a flat "N prompts per day" number stops meaning anything. Ten quick questions and ten Deep Research reports are not the same load, and Google's own wording now reflects that.

Why third-party numbers are still everywhere

Articles that rank for "how many prompts per day Gemini" are often older posts that were never updated, or round-ups that copy each other. Some were accurate when written. The problem is that Google's page now says limits "may change without notice, including due to capacity constraints," so any fixed figure has a short shelf life even when it was once correct.

How to treat any number you find

1. Check the date on the article. Anything from 2025 describes a different system.
2. Check whether it names a source on a Google domain. If not, treat it as an observation, not a rule.
3. Test it against your own account: the Usage view in the app is the only number that applies to you.

If you want a broader comparison of how the big assistants differ for real work, our ChatGPT vs Claude vs Gemini guide for professional work covers strengths rather than caps.

Google's help page gives one sentence on this: "Your limit refreshes every 5 hours until you reach your weekly limit." Read carefully, that describes two layers.

Layer one: the 5-hour window

You have a block of usage that refills about every five hours. If you spend it quickly, for example with a few Deep Research runs, you wait for the next refresh. If you spend it slowly, you may never notice it exists.

Layer two: the weekly limit

The refreshes do not continue forever. They continue "until you reach your weekly limit." After that, waiting five hours is not enough. The page says that if you hit the five-hour or weekly limit you can upgrade or wait for it to refresh.

What Google does not tell you

Because of those gaps, I would not trust a blog that tells you "your week resets Monday at midnight." If Google has not published it, the author is guessing or reporting one account's experience. See the last section for how to look at your own status.

Practical reading of the two layers

• Capped within a single heavy session: expect relief at the next 5-hour refresh.
• Capped repeatedly across days: you are probably nearing the weekly limit, and short breaks will not fix it.
• Capped on one feature only, such as Deep Research or video: the feature is expensive, and your general chat may still work.

The multipliers and context sizes below come from the Google help page. Google labels the lowest tier "Without an AI plan," and the subscriptions page describes the paid tiers as higher than the free plan. Prices vary by country, so I leave them out; check the page for your region.

What "4x" does and does not mean

It means four times the free baseline, and the baseline is not a published number. So "4x" is a ratio, not a promise of a specific count of prompts. If your free allowance runs out by lunchtime, AI Pro should stretch it about four times as far for the same kind of work. It will not remove the weekly limit.

Context window matters as much as the multiplier

A 32k-token window is roughly the size of a long article or a modest document. A 1M-token window is large enough for book-length material. If your work is long documents and not volume of questions, the context difference may matter more to you than the usage multiplier. On the other side, long context is also what drains your usage fastest, which the next section explains.

Where Google is vague

The subscriptions page and the help page do not always line up on model names per tier, and the subscriptions page says you can raise your cap by buying AI credits. The help page does not explain credits in detail. If credits matter to your decision, read Google's current terms for them before paying.

Ultra's two tiers

The help page lists Ultra as "5x or 20x higher than AI Pro limits depending on your subscription." That is the only description I found. It is a niche tier, so if you are comparing Plus and Pro, you can skip it.

Google's help page ranks requests by usage cost. I am quoting its categories, then translating them into habits.

Long conversations are the quiet cost

The page lists chat length as a factor. In practice that means a 60-message thread where every reply carries all the earlier context costs more per message than a fresh chat. When you are working on a new sub-task, paste a short summary into a new conversation instead of continuing the old one.

Deep Research and media generation

Google says these features "will consume more of your usage." Run a Deep Research report only when you will use the output. For a quick fact check, a normal prompt, or a web-grounded one through a tool like web search, is cheaper in every sense.

Model choice

The page says the models you use affect usage. It does not publish a per-model rate. The sensible rule is to use the lightest model that does the job and move up only when the answer is not good enough.

A habit checklist

• Start a new chat for each new task.
• Ask for the answer first and the long version second.
• Attach a section of a document, not the whole file, when a section will do.
• Batch related questions into one prompt instead of ten.

According to the help page, subscribers can "continue your conversation with Flash-Lite" after reaching a limit. If you hit the five-hour or weekly limit, you can upgrade or wait for the limit to refresh. Free users may find heavy features like Deep Research unavailable during high demand.

What you will likely see

A message that you have reached your limit, usually with some indication of when you can try again. Exact wording changes, so I will not quote a banner I cannot verify. If the message does not give a time, assume the next 5-hour refresh unless you have been capped for several days.

Free users can be restricted first

The page says that if capacity drops, users without a plan may be restricted before users with a plan. That is a reason you can be blocked on a busy day even if you have barely used the app.

How to check where you stand

1. Open the Gemini app and look for usage or limits information in the app's settings or the message itself.
2. Open the Google help page above and compare what your plan says to what you see.
3. If the numbers do not match, trust the app and note the date.

I am deliberately not describing menu names, because Google changes them and I cannot verify them from the pages I read.

Is it a limit or an outage?

If you are blocked very early in the day with little usage, check whether Gemini is slow or erroring for other people. High demand and capacity limits are separate from your personal usage, and the help page mentions both.

Paying fixes this only if the cause is volume. Switching fixes it if the cause is a single heavy feature or a single provider's cap.

Pay if these are true

• You are capped across several days, not just one long session.
• Your work needs long context, such as large documents or codebases.
• You already use Gemini for most of your tasks and like the results.

Switch if these are true

• You hit the cap once or twice a month, on a few heavy tasks.
• Your task is generic: drafting, rewriting, summarizing.
• You already pay for another assistant that is idle.

A simple decision table

The cost of guessing

Because Google does not publish the base number, no one can tell you in advance that "Pro will be enough for you." Use a month of honest tracking: note when you were capped and what you were doing. If heavy features caused most of it, the multiplier may not help as much as changing habits.

None of these are Google-published tricks. They follow directly from the factors the help page lists: prompt complexity, models and features, and chat length.

1. Put the answer you want in the first line

Vague prompts produce long answers and follow-up rounds. A specific first prompt often removes two or three exchanges.

Summarize the attached contract in 5 bullets. Flag any clause about [termination / auto-renewal / liability]. Do not rewrite the clauses. Keep it under 150 words.

2. Restart long threads with a handoff note

When a thread passes a few dozen messages, ask for a summary and start a new chat with it.

Summarize what we have decided so far in under 200 words: goals, constraints, decisions made, open questions. I will paste this into a new chat.

3. Trim attachments

Google lists large file analysis as medium usage. If you only need pages 12 to 20 of a report, upload those pages. This also helps on the free tier, where the context window is 32k tokens.

4. Batch related questions

One prompt with five numbered questions is usually cheaper than five prompts, because the shared context is processed once.

5. Save heavy features for the end

Run Deep Research or media generation when you have finished your lighter work, so a refresh is closer by the time you need chat again.

6. Do the search part somewhere else

If you are using Gemini mainly to find and quote web sources, a prompt grounded in search is a lighter task than a full research run. Our web search feature is built for that kind of lookup.

7. Split work by provider

Draft in one model, check in another. That keeps any single cap from becoming your bottleneck and also catches mistakes, since two models disagree in useful ways.

Mistake 1: Trusting a number from an old article

If a post says you get a fixed count of prompts, check its date and its source. The current Google page does not give one.

Mistake 2: Paying before measuring

A multiplier of 2x or 4x is relative to a baseline Google does not publish. Track what caps you for two weeks before you subscribe, then decide.

Mistake 3: Assuming the cap means a bug

The help page states that limits can tighten when capacity is constrained, with users without a plan restricted first. A block can be about system demand and not about you.

Mistake 4: Refreshing the page repeatedly

Reloading does not reset anything. If the message points to a refresh time, use that time to switch tasks or models.

What to tell a teammate who asks

Gemini's limit depends on how heavy your requests are, not just how many. It refills about every 5 hours until you hit a weekly cap. Google has not published the exact numbers, so check the help page and your own account instead of an article.

Most people hit the cap in the middle of a task, which is the worst time to learn a new tool. Here are the options in order of effort.

1. Shrink the context and retry

Open a new chat, paste the minimum you need, and ask for the specific output. This often works when the problem is a very long thread.

2. Switch to a different model for this task

Gemini's cap does not affect ChatGPT, Claude, Grok, DeepSeek or Mistral. If you use the Copilotly extension, you can change to another model from the same sidebar without copying your text between tabs. See the Gemini page, the Claude page or the ChatGPT page for what each model is good at.

3. Use your own API key

If you have a Google API key, the bring-your-own-key feature lets you use it in Copilotly. Be clear about what that means: a Google API key has its own separate quota and is billed by usage, not by the Gemini app's plan. It is not unlimited, and it can cost more or less than a subscription depending on how much you send.

4. Wait for the refresh

If you are under the weekly limit, the next 5-hour refresh should restore usage. Use the time on work that does not need an assistant.

A copyable plan for the next time it happens

When Gemini caps me:
1. Save the last good answer and my prompt in a note.
2. Open the same task in [backup model] with a one-paragraph summary of the context.
3. Re-run only the step I was on, not the whole conversation.
4. Note the time and what I was doing so I can see patterns over the week.

Our guide to using your own API key goes into costs in more depth.

Last verified: November 11, 2026. I read the Gemini Apps Help page on usage limits and the Gemini subscriptions page on that date. The subscriptions page opened in an Arabic, United Arab Emirates version for me, so I used it only for the compute-based wording and the refresh sentence, not for prices.

How to re-check in two minutes

1. Open the help page and read the paragraph beginning "Your limit refreshes."
2. Compare the multipliers and context windows to the table above.
3. Look for any new mention of a weekly number, a daily number or a new plan.

If something conflicts

Trust Google's page and your own app over this post. If you spot a difference, the likely cause is that Google updated the page, since it says limits may change without notice.

Related guides

For the same question on other assistants, see our ChatGPT free vs Plus limits guide and the Claude usage limit guide.