Gemini Usage Limits (2026): What Free, AI Pro, and AI Ultra Actually Cap, and When They Reset

Gemini usage limits: the short answer

As of September 2026, Google publishes no prompt count for any Gemini plan. What it publishes is a compute-based limit that refreshes every 5 hours until you hit a weekly cap, and a set of multipliers: the free tier gets "standard limits," Google AI Plus ($4.99) gets 2x, Google AI Pro ($19.99) gets 4x, and Google AI Ultra gets 5x AI Pro at $99.99 or 20x at $199.99. Media generation, Deep Research, the Pro model, and Deep Think burn the allowance fastest, and paid users who hit the wall can keep chatting on Flash-Lite. Everything below comes from Google's Gemini subscriptions page and the Gemini Apps help center, read on September 15, 2026.

If you came here looking for "50 prompts a day," that number stopped existing on May 17, 2026, when Google moved Gemini to compute-based metering. Here's what replaced it. The same question for every other vendor, side by side, is on AI usage limits by plan.

Gemini limits on Free, AI Plus, AI Pro and AI Ultra

Two things the table can't show. First, the multipliers are relative to a "standard limit" Google never states in absolute terms, so 4x is a real ratio applied to an unknown base. Second, all three models (Flash-Lite, Flash, and Pro) are available on every tier including Free; what the paid tiers buy is more compute against them, not access to them. My verdict on whether the $19.99 tier earns its keep is on is Gemini worth it.

The per-unit math Google's ladder hides

The unknown base cancels out if you price everything in it. Call the free tier's allowance one unit. AI Ultra is sold as a multiple of AI Pro, and AI Pro is 4x Free, so Ultra 5x is 20 units and Ultra 20x is 80.

So the cheapest Gemini, by allowance, is the $4.99 tier, tied with the $199.99 one. AI Pro and Ultra 5x charge twice as much per unit. That doesn't make AI Pro a bad buy, because the $15 gap mostly pays for other things: the 1M context window instead of 128K, 1,000 Flow credits instead of 200, and 5 TB of storage instead of 400 GB. It does mean that if your only complaint on Free is running out, AI Plus is the upgrade that fixes exactly that at the lowest price per unit. The assumption: Google's multipliers apply to the whole allowance, the 5-hour refresh and the weekly cap alike, which its table implies but doesn't spell out.

How the Gemini meter works

Google's help center describes the limit in one sentence: "compute-based usage limits that determine how much you can interact with Gemini tools and features. These limits factor in the complexity of your prompt, the models and features you use, and the length of your chat." Three consequences follow.

• The model you pick sets the burn rate. Flash-Lite is the "efficient workhorse," Flash balances speed and reasoning, and Pro is "our most advanced model" with longer response times. Google says outright that "more advanced models and higher thinking levels consume more of your usage."
• Chat length compounds. A long thread costs more per turn than a short one, because the whole conversation is context. Starting a new chat for a new topic is the cheapest habit on this page.
• Features are priced in compute too. Google lists the expensive ones: image, video, and music generation, Deep Research, the Pro model, and extended thinking or Deep Think. Deep Think is Ultra-only and requires the Pro model, so it's the most expensive thing you can ask Gemini to do.

You can see where you stand at gemini.google.com under Settings, then Usage Limits. Gemini also notifies you when you're close, and again when you hit it, with the refresh time.

The 5-hour refresh and the weekly cap

Gemini uses the same two-layer structure OpenAI and Anthropic use, and Google's wording is the clearest of the three: "Your limit refreshes every 5 hours until you reach your weekly limit." So there are two ceilings. The 5-hour one refills on its own during the day. The weekly one sits above it, and once you've spent that, the 5-hour refresh stops helping until the week rolls over. Google doesn't publish how large the weekly limit is on any tier, or exactly when it resets, so I'm not going to invent either.

The practical read: a Pro subscriber who runs three Deep Research reports and a video generation before lunch can hit the 5-hour ceiling, wait, and be fine by mid-afternoon. A subscriber who does that every day will meet the weekly ceiling instead, and that one doesn't refresh at 3pm.

Deep Research, video, and images

None of these carries a published count. Google's feature table shows Deep Research available on every tier including Free, video generation on the paid tiers, and image generation with Nano Banana 2 everywhere, but the only quantitative statement is that they "consume more of your usage." Two tier-specific notes do appear. For users without a Google AI plan, "some features that require more compute resources, like Deep Research, may be unavailable during periods of high demand." And when capacity changes, "limits for users without a Google AI plan may be limited before users with a plan." Free is first in line to lose Deep Research on a busy day.

Video and image creation outside the Gemini app, in Google Flow, run on a separate credit system that Google does quantify: 200 Flow credits on Plus, 1,000 on Pro, 10,000 or 25,000 on Ultra. That's the one place in Google's plan lineup where a hard number exists.

The context window is the published number

Google is precise about the thing it can measure. The free tier's context window is 32K tokens, AI Plus is 128K, and AI Pro and Ultra are 1M, which Google translates as "up to 1,500 pages of text or 30,000 lines of code." Exceeding the window doesn't stop the chat; it means Gemini stops seeing all of what you gave it, and Google warns responses may "miss connections or details throughout the content." For anyone uploading a contract or a codebase, the jump from 32K to 1M is the real reason to pay $19.99, more than the 4x multiplier.

Google only gives the pages figure for 1M. Scaling it down in proportion, which is my arithmetic and not Google's, gives a rough guide to which plan your documents need:

On that scale a 60-page contract is already more than Free can hold at once, and a 300-page document set needs Pro.

Hit the Gemini limit? Flash-Lite, AI credits, or wait

• Subscribers: "you can continue your conversation with Flash-Lite." The chat keeps going on the cheapest model until your limit refreshes.
• Anyone: upgrade to a higher plan, or wait for the refresh time Gemini shows you.
• AI credits: the subscriptions page says you "can extend your limits by purchasing AI credits." Google doesn't publish the price per credit on the pages I read.
• Limits can move. Google says "limits may change without notice, including due to capacity constraints." Plan for the multiplier, not for a fixed quota.

If you keep hitting the app's caps for work that runs the same way every time, the API bills per token instead of resetting on the app's five-hour window (it has its own rate limits). Gemini 3.8 Flash is $0.75 in and $3.75 out per million tokens until December 31, and the free tier's catch is that Google may use what you send to improve its products. The numbers are on the Gemini API pricing page.

Gemini vs ChatGPT on limits

The structures are close and the disclosure isn't. ChatGPT treats plain chat as unlimited on every plan and meters only reasoning, GPT-6 Pro and its agent work, and for those OpenAI prints actual per-window ranges. Gemini meters everything, chat included, and Google prints only ratios. If you value knowing your number, ChatGPT tells you more; if you value a $4.99 tier and a 1M context window at $19.99, Gemini is the cheaper ladder. The full ChatGPT side is on ChatGPT usage limits, and the head-to-head on everything else is on Gemini vs ChatGPT.

Five habits that stretch a Gemini allowance

1. Default to Flash. Reserve Pro for the tasks that need reasoning across a long document, and Deep Think for almost nothing.
2. New chat, new topic. The single biggest lever, since chat length is one of the three inputs Google names.
3. Batch Deep Research. Run reports in one block early in a 5-hour window, so the refresh lands before you need the next set.
4. Watch the weekly, not the 5-hour. Check Settings, then Usage Limits mid-week. The 5-hour ceiling is a pause; the weekly one is a stop.
5. Business accounts: don't assume the multipliers. Workspace and Gemini Enterprise publish no limit. Test with one seat for a month before rolling it out.

Sources

• Google, Gemini subscriptions (plan prices, 2x / 4x / 5x / 20x, the compute-based footnote, Flow credits, storage, 1,500 pages): read September 15, 2026
• Gemini Apps Help, limits and upgrades for Google AI subscribers (usage table, context windows, Flash-Lite fallback, Settings > Usage Limits, capacity wording): read September 15, 2026
• Gemini Apps Help, compute-based usage limits (the May 17, 2026 change, which features burn faster): read September 15, 2026
• Gemini Apps Help, continuing with Flash-Lite: read September 15, 2026
• Google Workspace pricing (Business Starter, Standard, Plus; Gemini Enterprise Business edition): read September 15, 2026