Blog · AI tools · Published
Gemini Usage Limits and Upgrades: How the Free Tier and Google AI Plans Differ
The free version of Gemini calculates limits based on compute consumption, taking into account prompt complexity, the model and features used, and conversation length; limits refresh every 5 hours until the weekly limit is reached. The context window for unsubscribed plans is 32,000 tokens. Google AI Plus and AI Pro raise the limit to 2 times and 4 times the standard limit respectively; AI Ultra is 5 times or 20 times the AI Pro limit depending on the specific subscription, with windows rising to 128,000 and 1 million tokens. After reaching the limit, subscribers can continue the conversation with Flash-Lite; the official approach for unsubscribed users is to wait for the refresh or upgrade. These figures apply to personal Google accounts; work or school accounts follow separate official guidance. The article lays out a four-tier comparison, the usage patterns that consume the most, where to check usage, and three checks to make before upgrading.
The free version of Gemini does not have a fixed "X times per day" quota. Instead, it is a usage pool calculated by compute (computing resource) consumption: the official documentation states that limits take into account prompt complexity, the model and features used, and conversation length, refreshing every 5 hours until the weekly limit is reached (Google official help page). Upgrading to a Google AI plan buys higher usage multipliers and a larger context window, not the removal of limits. People searching for 「gemini 3 免費」 are really trying to decide two things: whether the free version can support the way they use it, and whether to pay for more usage; the multipliers and window figures below are officially stated to apply to personal Google accounts, while work or school accounts follow separate official guidance.
1. What the free tier actually limits
The official name for this mechanism is "compute-based usage limits." It measures not just how many messages you send, but how much computing resource those requests consume in total. The variables involved include prompt complexity, the model and features used, and conversation length.
Refreshing happens on two cycles: every 5 hours, until the weekly limit is reached. If you use up the short window, you wait for the next 5 hours; if the weekly total is used up as well, you wait for the weekly limit to refresh. So "can I still use it today" depends on which window you are in and how heavily you have used it, not on a fixed number of messages.
The context window (the total amount of content the model can read at once; officially compared to Gemini's "reading ability") is a separate line. Once the prompt content exceeds the window, the official documentation states this may prevent the model from fully considering all the content you provided, or cause it to miss connections and details between pieces of content — more noticeably when details are spread across several large files.
There is also one point directly relevant to free users: officially, users without a subscription may find that certain features requiring more computing resources (for example Deep Research, which performs multiple rounds of search before reasoning and generating a report) are unavailable during peak usage periods.
2. Where the numbers differ across the four plans
The multipliers and window figures below are stated on the official page to apply to users managing the Gemini app through a personal Google account; if you use a work or school account, the rules follow the official guidance for that type of account and should not be copied over. In the table, token is the unit the model uses to process text, roughly a short segment of text.
| Plan | Usage limit (official label) | Context window | After reaching the limit |
|---|---|---|---|
| No AI plan subscription | Standard limit | 32,000 tokens | Wait for the 5-hour window or weekly limit to refresh, or upgrade to a Google AI plan for higher limits |
| Google AI Plus | 2 times the standard limit | 128,000 tokens | Can continue the conversation with Flash-Lite; can also wait for a refresh or upgrade |
| Google AI Pro | 4 times the standard limit | 1 million tokens | Can continue the conversation with Flash-Lite; can also wait for a refresh or upgrade |
| Google AI Ultra | 5 times or 20 times the AI Pro limit, depending on the subscription plan | 1 million tokens | Can continue the conversation with Flash-Lite; can also wait for a refresh or upgrade |
Officially, for example, a 1 million token window can understand up to 1,500 pages of text or 30,000 lines of code.
Three things to note when reading the table.
First, the multiplier is the size of the same usage pool, not the length of a feature list. The official wording is "2 times the standard limit" and "4 times", not "unlimited"; a higher allowance does not mean there is no longer a cap.
Second, the Ultra row reflects two official labels for different Ultra subscription plans. Which one actually applies depends on the plan shown on your account's subscription page; this article does not treat 5 times and 20 times as a range you can choose from.
Third, feature availability is a separate line: officially, some features are available only to Google AI subscription users, and some features (such as media generation or Deep Research) consume more usage. If what you lack is the feature itself, simply raising the multiplier will not solve it.
3. Which usage patterns consume the most
Officially, common prompts are grouped into five tiers by "relative usage cost", which explains the issue better than memorizing a count:
| Prompt type | Relative usage cost | Official description and examples |
|---|---|---|
| Simple exchanges in a conversation | Very low | Short text prompts without follow-up questions; e.g., help me write an email subject line |
| Ongoing conversations | Medium | Requires re-reading the earlier conversation to stay coherent; the longer the conversation, the more it consumes; e.g., long conversations with dynamic edits, coding projects that need repeated revisions |
| Large file analysis | Medium | Long documents, codebases or spreadsheets require reading large amounts of data; e.g., analyzing a 100-page PDF financial report |
| Deep Research | High | Requires multiple multi-step searches and multiple rounds of reasoning; e.g., a deep competitive analysis of a particular area |
| Video generation | High | Multimedia requests consume large amounts of compute; video generally requires the most |
From this, three directly actionable practices can be drawn.
- Start a new conversation when you change topic. The official reason is specific: this avoids re-reading a lengthy conversation history before generating an answer, which helps save usage.
- Start a new thread when a conversation goes off track. Ongoing conversations carry a medium relative cost, and every turn re-reads the earlier text; copying existing conclusions into a new conversation and continuing there is more economical than repeatedly correcting an old one.
- Split large files before asking. If you often paste long documents or code in one go, what is holding you back may not be the number of messages but how much can be read at once; split the material by chapter or module and ask in sections first, confirm that the current window is genuinely insufficient, and only then consider paying for a larger window.
4. How to check usage, and the options after reaching the limit
The official path for checking is three steps: open gemini.google.com; in the lower left, select the "Settings" icon and then the "Usage limits" icon; view the usage limit for your current plan.
Officially, the app notifies you when you approach the limit, and sends another notification after you reach it, stating when the limit will refresh. The basis for judgment is the prompt actually displayed in your account, not counting how many questions you asked today yourself.
After reaching the limit, there are three officially stated options:
- Wait: wait for the 5-hour window or weekly limit to refresh before continuing.
- Upgrade: upgrade to a Google AI subscription plan for a higher usage limit.
- Switch: users subscribed to a Google AI plan can continue the conversation with Flash-Lite after reaching the usage limit.
The third option exists only after subscribing; for users without a subscription, the official guidance is to wait for the limit to refresh or upgrade the subscription plan.
5. Whether to upgrade: three checks
Don't rush to compare tiers first. Run through these three steps, and you can rule out some cases where an upgrade is not actually needed.
Step one: identify which type of usage is holding you back. Officially, short exchanges are listed as "very low" relative usage cost, ongoing conversations and large file analysis as medium, and Deep Research and video generation as high; compare this with what you most often do, and you will know where the limit is mainly being spent.
Step two: check whether the context window is insufficient. The boundaries for the three windows are in the table in section two; if you often need to paste very long documents or code in one go, first split the material and ask in several turns, confirm that a larger window is genuinely needed, and only then consider upgrading.
Step three: check whether the feature you want is included. Some features are available only to subscription users, and media generation and Deep Research also consume more usage; if what you lack is the feature itself, raising the multiplier will not help.
Once all three steps are done, go back to the table to match the tier: Plus is 2 times the standard limit, 128,000 tokens; Pro is 4 times, 1 million tokens; Ultra is 5 times or 20 times the AI Pro limit, 1 million tokens. What the official documentation provides is relative multipliers, not counts, so do not convert the multipliers into "how many questions per day" and then base a decision on that.
There are two more conditions that are easy to misread before upgrading. First, the limits themselves can change: officially, limits may be adjusted without notice, for example when system load is excessive, and may also be adjusted when activity increases significantly. Second, a subscription does not necessarily always come first: officially, if load levels change, users subscribed to a Google AI plan may hit limits earlier than users without a subscription. Both are official statements of conditions and should not be extended to mean "paying equals priority".
There are also two hard conditions related to whether you can use the service at all, unrelated to quota: in most countries/regions, you must be at least 13 years old (or the locally required age) to use the Gemini app through a Google AI plan; in the European Economic Area, Switzerland or the UK you must be at least 18, and those under that age may not be able to use certain features; Ultra is available only in countries/regions where it has launched, and when traveling from a launched region to one where it has not launched, you may still see related notices but cannot use or upgrade until you return to a supported region. This article follows the officially stated conditions and does not involve any method of bypassing regional restrictions.
6. How this relates to our business
This article only answers questions about usage and upgrades for the Gemini app. Our site handles ChatGPT top-ups; the quota rules and after-sales rules on the two sides cannot be applied interchangeably (ChatGPT top-up plans, top-up service terms). If what you want to compare is how to choose among ChatGPT plan tiers and how costs are calculated, ChatGPT subscription cost overview is more direct.