Since May 2026 the Gemini app has had compute-based limits. Google no longer counts prompts; it weighs each one by how complex it is, which models and features it uses, and how long the chat already is. Your allowance refreshes every five hours until you reach a weekly limit. Google AI Plus gets twice the free limit, AI Pro four times, and AI Ultra five or twenty times AI Pro. If you're hitting it, the cause is almost always one of four habits, and Google's own list of what it counts says as much.
How do Gemini's usage limits work?
Two clocks. A five-hour limit that refreshes on its own, and a weekly limit that stops the refreshing once you reach it. Gemini warns you when you're close and tells you when the limit will refresh once you hit it. You can check where you stand at gemini.google.com/usage.
The allowance is per account, not per device. The web app, the phone app and Gemini in Chrome all draw on it.
When you run out, you can wait or upgrade. Google has also said paying subscribers drop to a smaller model rather than being cut off, and that AI Pro and Ultra subscribers will be able to buy credits to keep going.
What is Google actually counting?
Google names three things, and they line up with how we explain token use for every assistant. We call it Intersect's Four Ls.
What Google counts, and the L it maps to
| Google counts | What that means | Intersect's Four Ls |
|---|---|---|
| The length of your chat | Every reply carries the whole conversation | Length |
| The complexity of your prompt | Long pastes, big files, connected data | Load |
| The models and features you use | Pro, Thinking, Deep Think, Deep Research, images and video | Lift |
| How often you ask again | Not on Google's list, but every retry is all three again | Loops |
Chat length is the one people miss. Each time you send a message, Gemini reads the conversation above it again. Here's what that does to the same thirty questions asked two ways:
Same answers. About 60% less reading.
One long chat makes Gemini read about 279,000 tokens. Three short chats with a handover note between them, about 107,000. That's roughly 60% less work for the same answers, and Google counts the work.
Intersect's Four Ls
How long the chat has run.
Google counts chat length directly. Each reply costs more than the last.
One task per chat. Move on with a handover note.
Everything in the prompt besides your question.
Long pastes, big files and connected apps make each prompt more complex.
Send the part, not the whole file. Split big jobs.
The model and features you pick.
Pro, Thinking, Deep Research, image and video generation count for more than Fast.
Fast by default. Step up per job.
The rounds it takes to get it right.
Every retry repeats the length, the load and the lift.
Brief properly. Edit the prompt instead of correcting it.
How do you cut the length?
One task, one chat. When the task changes, start again, and carry over only what matters with a handover note:
We're moving to a new chat. Write a handover note I can paste into it. Include: the goal, what we've decided and why, the latest version of the work (or where it lives), anything still open, and the next step. Leave out anything we tried and dropped. Keep it under 300 words.
How do you lighten the load?
- Send the part, not the pile. Ask about one tab of the spreadsheet or the clause you care about, not the whole file.
- Split big jobs. Five separate questions about five documents cost less than one prompt carrying all five.
- Don't paste what Gemini already has. If the document is in the chat, refer to it. Pasting it again doubles it.
How do you match the lift to the job?
Google lists Pro, extended thinking, Deep Think, Deep Research and image and video generation among the features that count for more. Fast is the cheapest way to get most everyday work done. Here's how we'd set it:
Match the lift to the job
| Job | Mode | Feature |
|---|---|---|
| Rewrite an email, summarise a page, quick answer | Fast | None |
| Draft a proposal or report from your notes | Fast, or Thinking for structure | None |
| Analyse a spreadsheet, contract or model | Pro or Thinking | File upload |
| A sourced report across many sites | n/a | Deep Research |
| A hard maths, science or strategy problem (Ultra) | Deep Think | None |
| Images or video | n/a | Only when you'll use the result. These are the heaviest features |
Watch the newest models too. When Gemini 3.5 Flash arrived in May, users reported single prompts taking several percent of their five-hour allowance. A model that's better isn't always cheaper.
How do you cut the loops?
- Brief it properly. Audience, length, purpose and what good looks like, in one message.
- Batch related questions rather than sending them one by one.
- Edit, don't correct. Edit your prompt and Gemini answers the new version, instead of carrying the wrong answer forward.
Where does this break?
Google doesn't publish a number, so you find the edge by reaching it. Use the usage page to learn what your normal week looks like before you plan a big job around it.
Some work needs the long chat. If you're revising one document against comments from earlier in the conversation, splitting it loses the thread. Keep it together and accept the cost.
And if your team runs Deep Research or video most days, the habits only go so far. That's what AI Ultra is priced for.
Six changes, ten minutes
- Open gemini.google.com/usage and note where you stand.
- Start a fresh chat for your next task. Bring anything you need over with the handover prompt.
- Upload the section you're asking about, not the whole file.
- Use Fast for routine jobs and step up to Thinking or Pro only when a question needs it.
- Keep Deep Research, images and video for work you'll actually use.
- Next time Gemini misreads you, edit your prompt instead of correcting it.
Want your team getting more from the same plan?
We help teams set up Gemini, ChatGPT or Claude properly: the right plan, the right settings for each kind of job, and an hour with your people on habits like these. If you're hitting limits every week, we'll tell you whether it's the habits or the plan.
FAQs
How many prompts do I get on Gemini?
There's no fixed number any more. Since May 2026 limits are compute-based, so a short question on Fast costs far less than a Deep Research report.
Does a new chat reset my limit?
No. It resets what Gemini has to re-read. The five-hour and weekly limits run on their own clocks.
Do the phone app and the web app share a limit?
Yes. Your allowance is per account, across devices.
What happens when I hit the limit?
Gemini tells you when it refreshes. Paying subscribers can keep going on a smaller model, and Google has said credits will let AI Pro and Ultra subscribers extend their limits.
Usage limits, assistant by assistant
Related reading
- Hitting Your Claude Limit? Here's Where the Tokens Go
- ChatGPT Says Unlimited. So Why Did You Hit a Limit?
- Set Up a Shared AI Workspace for Your Team in an Afternoon
Sources
- Gemini Apps Help: Gemini Apps limits and feature availability, accessed 6 November 2026
- 9to5Google: Gemini's new usage limits are live, 21 May 2026
- 9to5Google: What you get with Google AI Plus, Pro and Ultra, May 2026
- PCWorld: Google just made big changes to Gemini usage limits, May 2026
- ARY News: How Gemini's new compute limits affect heavy users, May 2026
- AI Toolbox: Gemini usage limits 2026, reviewed October 2026