Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- 429 does not only mean requests are too frequent. Read the error body to determine whether the issue is RPM, TPM, RPD or a temporary capacity shortage.
- Gemini quotas are calculated per Google Cloud project, not per API key. Creating another key in the same project generally does not give you a new quota.
- Use a bounded number of exponential-backoff retries with random jitter only for transient errors such as 408, 429 and 5xx. Do not blindly retry 400 or 403.
- RPD resets at midnight US Pacific time. Preserve the time zone, model, project, status code and request ID when investigating.
- If a leaked-key warning appears, revoke and rotate the key immediately. Never put real keys in browsers, repositories, screenshots or tickets.
First identify quota, capacity, permission or network failures
When Gemini API returns an error, save the HTTP status code, JSON error body, model name, request time, Google Cloud project and request ID first. 429 usually relates to quota or service capacity; 503 often means temporary unavailability or insufficient capacity; for 403, prioritize the key, project, region and permissions. DNS, connection-timeout, TLS and proxy-authentication failures belong to the network layer.
Do not change proxies or keys based solely on a generic call-failed message. Reproduce the issue with a minimal request containing no sensitive data, then use the API Key, Base URL and Model configuration guide to confirm the endpoint and model identifier.
What RPM, TPM and RPD limit
RPM is requests per minute, TPM is input tokens per minute, and RPD is requests per day. Exhausting any dimension can trigger a limit, so a low request count does not rule out exceeding TPM. Long contexts and concurrent batches are especially likely to hit the token dimension first.
Official documentation states that quotas apply per project, not per API key. Splitting a project's requests across multiple keys does not increase its total quota. RPD resets at midnight US Pacific time. Teams across time zones should record both local time and that reset basis in monitoring and tickets.
Which errors can be retried
For 408, 429 and 5xx, use exponential backoff only after confirming the request is idempotent or safe to replay. For example, wait about 1, 2, 4 and 8 seconds, add random jitter, and set maximum attempts and total waiting time. If the response includes Retry-After, prioritize the server's guidance.
400 usually indicates parameters or a request-body issue; 403 generally points to permissions, keys or policy boundaries. Replaying the same request does not fix them. Do not retry persistent 429 responses indefinitely either. Reduce concurrency, shorten input, queue requests or apply for a higher quota to avoid a retry storm amplifying costs and the incident.
A minimal troubleshooting flow for 429 and 503
First, check the error body and quota console to identify the triggered dimension. Second, reduce concurrency to 1 and send a minimal request with short input and the same model. Third, observe whether bounded backoff restores service. Fourth, compare another already-authorized model or an off-peak period, but do not switch accounts to evade limits. Fifth, record results before adjusting queues, caching and concurrency.
If all models fail before a connection is established, or you see DNS, TLS, 407 or connection-timeout errors, use the proxy connection troubleshooting checklist. Network egress can improve connectivity, but cannot increase Gemini project quota or lift a key ban.
Handling a leaked-key warning
If the error explicitly says the API key was reported as leaked, stop using it, revoke or delete it in the Google Cloud console, create a restricted replacement, and inspect repository history, frontend bundles, CI logs, screenshots and tickets for exposure. Retrying or changing network egress cannot restore a blocked key.
Store the new key only in server-side secrets management or controlled environment variables, restricting callable APIs, origins and permissions. Verify it with a minimal request after rotation before restoring traffic. Also remove complete Authorization values, keys and user data from logs.
Pre-release checklist
At minimum, check that RPM/TPM/RPD are monitored per project; different models have separate queues; retry count, total duration and jitter are bounded; 400/403 go directly to manual or configuration investigation; logs retain status codes, error codes, request IDs and models but no keys; cost and failure-rate alerts exist; and fallback paths do not cause duplicate charges or repeat side effects.
If your team needs stable access to overseas API documentation and consoles, visit the PuppyIP website to learn about fixed network egress. It addresses connection paths only and does not promise increased official quotas, model availability or account permissions.
Sources
Frequently Asked Questions
Does Gemini API 429 always mean too many requests per minute?
No. RPM, TPM, RPD or service capacity can be involved. Use the error body and project-quota page as the evidence.
Can creating multiple API keys in one project increase quota?
No, not on that basis. Official documentation states that quotas are calculated per project, not per individual API key.
Can I keep retrying Gemini 503 indefinitely?
No. Use bounded exponential backoff with jitter and a total waiting-time limit. Reduce load or stop if failure persists.
Why is automatic retry not recommended for 403?
403 usually points to permissions, keys, projects or policy boundaries. Retrying an unchanged request does not alter those conditions.
When does RPD reset?
Gemini's official documentation states that the daily request quota resets at midnight US Pacific time. Record accurate times when troubleshooting across time zones.
Can a proxy resolve Gemini 429?
A proxy affects only the network path and cannot increase project quota. Investigate the network only when evidence points to DNS, connection, TLS or proxy authentication.