Pause
Wait before retrying. If the limit is RPM or TPM, a short delay may be enough.
API rate limits
Gemini API limits are project, model and tier based.
For developers, the important limits are requests per minute, tokens per minute, requests per day, model-specific capacity, usage tier and sometimes spend-based limits. Check AI Studio for your active numbers.
Official model
How Google describes Gemini API rate limits.
Google’s Gemini API documentation says limits regulate requests in a timeframe, are evaluated across multiple dimensions, are applied per project rather than per API key, and can vary by model and usage tier.
| Dimension | What it means | What causes trouble |
|---|---|---|
| RPM | Requests per minute. | Sending too many concurrent or burst requests. |
| TPM | Input tokens per minute. | Long prompts, repeated context, large file text or many parallel requests. |
| RPD | Requests per day. | Heavy daily usage, retry loops, cron jobs or batch-like work through interactive endpoints. |
| Spend-based limit | Rolling spend protection on paid tiers. | Expensive model calls, large context windows or high output volume. |
429 fixes
What to do when you hit RESOURCE_EXHAUSTED.
A 429 is usually a capacity, quota or rate problem. Do not immediately retry in a tight loop.
Back off
Use exponential backoff with jitter. This prevents synchronized retry storms.
Reduce tokens
Shorten inputs, summarize context and lower output length.
Check tier
Open AI Studio to see active limits, billing status and upgrade eligibility.
Architecture
Design for limits before launch.
Put requests in a queue
Control concurrency centrally instead of letting every user action call the API immediately.
Cache repeated context
Do not resend the same long instructions or documents when a summary or cached context works.
Use model fallback
For low-risk tasks, route to a lighter model when the premium model is saturated.
Debug checklist
How to diagnose a Gemini API limit error.
Before changing code, identify which counter failed. A 429 can come from burst traffic, huge prompts, exhausted daily quota or billing constraints.
Log model and project
Rate limits are model and project dependent, so include both in your error logs.
Track prompt size
Record input token estimates and output length so TPM pressure is visible.
Separate retries
Mark retry attempts so they do not look like organic user demand.
Watch daily usage
If RPD is exhausted, short backoff will not help until quota returns.
Related developer limit guides