AI Error WikiSearch

Gemini API 429 RESOURCE_EXHAUSTED

A Gemini API 429 can involve request, token, daily or spending limits. Read the response details, check the model and project, then retry safely.

Quick answer

429 RESOURCE_EXHAUSTED means the Gemini API rejected this request under a limit or quota condition. A paid plan does not by itself prove this particular model, project, or time window has capacity. Read the response body and the project's current limits before changing code.

Exact message

429 RESOURCE_EXHAUSTED

Find which limit was hit

Record the model, project, response time, request volume, and any RetryInfo or quota fields in the error. Check the exact model and project's limits in AI Studio, including requests per minute, tokens per minute, requests per day, and any spend-based limit. The Gemini API rate-limit documentation explains that limits vary by tier and model; the API error reference separates per-minute limits, daily quota, and bursts. If a daily quota is exhausted, waiting seconds between retries will not reset the day.

Google's retry guidance

Google's troubleshooting guide recommends exponential backoff with jitter and a maximum attempt count for retryable 429 responses. Reduce concurrent requests or prompt/output size if they consume the relevant limit. An unlimited retry loop can keep hitting the same limit. If the response indicates a non-retryable billing or credential problem, resolve that condition instead.

What paid users reported

One paid-tier embedding user reported recurring 429s while believing usage was low. Another forum case reported a paid Tier 1 project whose displayed quota did not match the immediate error. These are individual reports; neither proves your project's quota is wrong. Preserve project ID, model, limit view, timestamp, and sanitized error for Google support if the numbers still conflict after checking the right model and project.

In the second case, a Google forum responder asked the user to submit details through the form linked in that thread after confirming no published rate limit was breached. That is the next supported escalation when the displayed limits and response still disagree; do not publish API keys or private request data in a forum reply.

Frequently Asked Questions

Why do I get 429 with paid credits left?

Credits and request limits are different checks. Inspect the model-specific RPM/TPM/RPD and spend window before concluding the quota display is inconsistent.

Should I just retry?

Use bounded exponential backoff for a transient rate limit. If the error identifies a daily quota or billing condition, fix or wait for that condition instead.

Is this the same as Gemini API 503?

No. The 503 guide covers service unavailability and deadline symptoms. Both may be retryable, but the reason and first checks differ.

Sources