Gemini API 503 errors appear to consume Free Tier RPD, leading to 429 quota exhaustion

Gemini API
1

I am reporting a reproducible issue observed with the Gemini Developer API using gemini-3.8-flash through Google Apps Script.

The integration uses controlled concurrency, a small number of tasks per execution, and an audit log. During 23-24 September 2026, Madrid local time, repeated HTTP 503 Service Unavailable responses were followed by HTTP 429 daily quota exhaustion.

Relevant audit entries include:

  • 2026-09-23 09:25 - HTTP 503
  • 2026-09-23 12:25 - HTTP 503
  • 2026-09-23 15:25 - HTTP 503
  • 2026-09-23 18:25 - HTTP 503
  • 2026-09-23 21:25 - HTTP 503
  • 2026-09-24 00:25 - one request completed successfully
  • 2026-09-24 03:25 - HTTP 429, with the response indicating the Free Tier limit of 20 requests per day

The broader audit contains multiple 503 responses. In one concrete case, the same queued task received repeated 503 responses across scheduled executions before eventually succeeding. This makes the interaction between service-side 503 failures and a small daily RPD allowance especially important.

Could Google please clarify whether requests returning HTTP 503 count against daily RPD; whether that is intentional when no useful model response is produced because the service is unavailable or overloaded; whether service logs for 23-24 September can be reviewed; whether there was a known incident affecting gemini-3.8-flash or the Developer API endpoint; and whether there is a recommended mechanism to prevent service-side 503 failures from consuming a significant portion of a small Free Tier daily quota?

As a defensive measure, I have changed the client policy so that a 503 causes no immediate retry, imposes at least a 12-hour defer, and uses a local hard budget of 19 API attempts per Google quota day, reserving one request.

I can provide additional audit timestamps or sanitized logs if useful. No API keys or credentials are included here.

Thank you for reviewing this behavior.

2

Hello @Hector_Herrera_Yuste ,

503 errors are due to our services being temporarily overloaded. Please see this pinned post for suggestions on how to handle this.

3

Hello Pooja,

Thank you for your reply. I understand that HTTP 503 errors can occur
when the Gemini API is temporarily overloaded.

However, this does not address the specific issue I reported. My
concern is not why the 503 errors occur, but whether requests that
fail with HTTP 503 are nevertheless counted against the Free Tier
daily request quota (RPD).

In my case, after a sequence of 503 responses, the API subsequently
began returning HTTP 429 quota-exhaustion errors. This raises the
possibility that failed 503 requests may have been counted toward the
daily RPD limit.

Could you please clarify explicitly:

  1. Does a request that returns HTTP 503 increment the Free Tier RPD counter?
  2. If it should not, what could explain the apparent RPD exhaustion
    following repeated 503 failures?
  3. Is this expected quota-accounting behavior, or could it indicate a bug?
  4. If it may be a bug, could this be escalated to the Gemini API team
    for investigation?

I am happy to provide timestamps, request logs, model details, or any
other diagnostic information that may help reproduce the issue.

I would particularly appreciate a direct clarification regarding quota
accounting for failed 503 requests, since the general guidance on
handling 503 errors does not resolve that part of the report.

Thank you,
Héctor Herrera Yuste