Find your Gemini API rate limits
For direct Gemini Developer API use, start with Google’s Gemini API rate-limits documentation. To check active limits, a Google AI Developers Forum reply dated December 8, 2025, links to the AI Studio rate-limit view and says it shows limits regardless of tier. Treat that as a forum-provided route, not confirmation that the current interface or project/model selection steps are unchanged; the thread does not explain those steps.
The documentation explains quota rules; check the active limits for the project and model you intend to call rather than relying on a generic tier estimate. Limits are applied per project, not per API key, so adding keys to the same project does not create separate quota pools. Limits vary by model, and some apply only to particular models or tools.
Which quota dimensions can apply?
Google says Gemini API limits are usually measured across these dimensions. Exceeding an applicable limit can trigger a rate-limit error.
| Dimension | Meaning | Planning note |
|---|---|---|
| RPM | Requests per minute | Smooth traffic across the project; bursts can exceed this even when token use is low. |
| TPM | Input tokens per minute | Large prompts can hit this limit before the request-count limit. |
| RPD | Requests per day | The documented reset is midnight Pacific time. |
| Other model or feature limits | Depending on the model, limits can include tokens per day (TPD), images per minute (IPM), or limits for built-in tools. | Check the specific model and enabled tools; these are not universal quotas. |
Experimental and preview models can have more restrictive limits. Google also lists spend-based limits per 10 minutes: Free: N/A; Tier 1: $10; Tier 2: $50; Tier 3: $200. Applicability and eligibility can depend on the project, so check the current documentation and active project limits rather than assuming these figures apply to every account.
Illustrative calculation—not a Google quota: if your AI Studio project view showed 40 RPM and 60,000 input TPM, and requests averaged 2,000 input tokens, the token limit would allow about 30 such requests per minute (60,000 ÷ 2,000). TPM would be the tighter constraint. Actual prompt sizes vary, so leave headroom and monitor traffic.
Keep API tiers separate from Google AI subscriptions
Google AI Pro and Ultra plan benefits for developer usage apply in the Google AI Studio web interface. Google says direct Gemini API use—such as API keys or external applications—is billed and managed separately. Do not assume a consumer subscription sets the production API quota for your project.
For API use, limits are associated with the project’s usage tier. Check the current project limits in AI Studio rather than treating a tier label as guaranteed throughput.
Some features have their own limits. For example, Google documents a default of 1,500 requests per day per project for Grounding with Google Search on Gemini 3 and newer models through the Gemini Developer API. This is a feature-specific limit, not the general generation-request quota.
If your app calls Gemini through Firebase AI Logic, do not treat its Firebase-specific quota as a direct Gemini API limit. Firebase documents a separate default of 100 Generate content requests per minute per user for calls through its SDK; the Gemini provider’s limits also apply and take precedence.
Handle a 429 in production
A 429 is not enough by itself to identify which limit was reached. Check any error details returned to your client and compare the request’s project and model with their active limits in AI Studio. The Gemini API troubleshooting guide recommends exponential backoff with jitter for specified transient errors, including 429; that does not mean every 429 is transient or that retrying will clear an exhausted quota.
Use the limit indicated by the available evidence when choosing a response:
- For RPM pressure, smooth or queue requests and limit concurrency.
- For input TPM pressure, reduce input context or pace requests. Shortening output does not directly reduce input-token use.
- For a spend-based limit, reduce request rate and, where appropriate, input context or output size. Check the current documentation and project view for whether the listed spend limit applies.
- For exhausted RPD, stop rapid retries and defer work; the documented daily reset is midnight Pacific time.
- For a transient, retryable error, use bounded exponential backoff with jitter and a maximum attempt count. Do not retry indefinitely. If a 429 persists without an apparent exhausted limit, stop after the bounded retries, preserve the returned error and the project/model context, recheck the active limits, and consult the troubleshooting guide’s 429 guidance. The available guidance here does not establish a separate support or reporting route for unexplained errors.
If normal production traffic repeatedly hits a confirmed limit, review Google’s rate-limit increase process. An increase is not guaranteed and does not replace managing request volume.
Sources
Rate limits | Gemini API | Google AI for Developers
Troubleshooting guide | Gemini API | Google AI for Developers
Google AI plans | Gemini API | Google AI for Developers
Rate limits and quotas | Firebase AI Logic
Tier 1 rate limits - Gemini API - Google AI Developers Forum