There isn’t one fixed Gemini API Free Tier limit that applies to every project and model. Google describes limits in several dimensions—including requests per minute (RPM), input tokens per minute (TPM), and requests per day (RPD)—and says they vary by model. Your usable allowance is the one shown for the project and model your application actually uses in AI Studio’s rate-limit view. Limits apply per project, not per API key. (ai.google.dev)
That means an old post quoting a particular number of daily requests is not a reliable budget for your project. Check the project’s active limits, then compare them with your expected calls and token use.
How to check your Gemini API Free Tier limit
- Open AI Studio’s rate-limit view and select the project used by your application.
- Find the exact model you plan to call. Limits vary by model; experimental and preview models can have more restrictive limits.
- Record every displayed quota dimension—for example, RPM, input TPM, and RPD, plus any model- or tool-specific limit shown.
- Check the project’s status and billing tier in AI Studio. A model’s free-price listing alone does not confirm that your project can use it on the Free Tier.
- Recheck after changing the project or model. Google says RPD quotas reset at midnight Pacific time. (
ai.google.dev)
The limits are separate checks: you can hit a per-minute request limit even if you have daily requests left, or run short of tokens while request counts remain available. Creating another API key in the same project does not create another project-level quota. (ai.google.dev)
Does a free API key mean free Gemini API use?
No. Creating a key, a model having a Free Tier price, your project being eligible, and your project having enough capacity are different things. Google’s pricing page describes Free Tier access as limited to certain models and subject to their free-tier rate limits. Google also says not all projects or accounts are eligible; use its troubleshooting guide if AI Studio shows a restriction or your request is denied. (aistudio.google.com)
Paid Tier 1 should be treated as paid API use, not as a free inference allowance. Google’s billing guide says linking billing moves a project to a Paid Tier for higher rate limits and paid features. Its troubleshooting guidance states that upgrading a restricted project to paid does not unlock Free Tier quota. Check the exact model and feature in the pricing table, and the project’s status and quota in AI Studio. (ai.google.dev)
Google’s billing guide says moving from Free to Paid involves linking a billing account and prepaying at least $5 USD (or the equivalent in another currency). Check the billing setup shown for your account before relying on that as your required starting amount. (ai.google.dev)
Estimate calls, tokens, and paid cost
First estimate the workload; don’t translate a quota into users or conversations without knowing how many model calls each interaction triggers.
Example (hypothetical): Suppose a small Q&A feature receives 100 visitor questions in a day. If each question triggers two model calls, that is 200 requests per day. If each call averages 1,200 input tokens and 300 output tokens, the daily total is 240,000 input tokens and 60,000 output tokens. These are workload assumptions, not a Google quota or a prediction of what your project will serve. Compare them with the project’s displayed request and token limits; also account for any bursts, retries, or additional model calls.
For paid usage, calculate input and output separately using the exact model’s current listed rates:
Estimated model cost = (input tokens ÷ 1,000,000 × input price per million) + (output tokens ÷ 1,000,000 × output price per million) + any separately priced tools or features.
Use the current official pricing row for the model, version, service mode, and features you intend to use. Prices can differ across models and features, so a rate from an older article—or from a different model—is not a sound budget assumption. The official pricing page lists the applicable pricing details. (aistudio.google.com)
Also count every model call in the application flow. For example, a local code-only filter can reject an off-topic message before it reaches Gemini; an LLM call to classify that message is an additional API request. The latter also needs to be included in token and cost estimates.
Is the Free Tier enough for your workload?
Use the Free Tier only if all three conditions hold:
- Your intended project is eligible and can make inference calls.
- The exact model and features you need have Free Tier pricing.
- The project’s active limits leave enough room for your expected request volume, token use, and traffic peaks.
If any condition fails, consider reducing unnecessary calls or tokens, choosing a suitable model with a free-price entry, or budgeting for paid usage. A paid tier offers paid access and higher limits, but your estimate should still use the applicable model prices and project limits—not assume unlimited capacity. (ai.google.dev)
A 429 RESOURCE_EXHAUSTED response can indicate that a limit was reached. Check the full error and the relevant project, model, and quota dimension before deciding what to change. Google recommends exponential backoff for errors that are appropriate to retry; repeated retries do not create additional quota. (ai.google.dev)
Bottom line
There is no single Free Tier request count to use as every developer’s budget. Check the intended project’s live quota for the exact model, confirm project eligibility and the model’s pricing, then compare those limits with your own call and token estimates. If you need paid access, budget using the exact current pricing row for your model and features.