429 RESOURCE_EXHAUSTED on Gemini API (paid Tier 1) — paid tier not enforced, requests stuck on free-tier quota

Project ID: restai-4725..

Billing account: 014999-3F9D97-1235.. (Tier 1, Postpay, active, valid card, no balance due)

Model: gemini-2.5-flash (generativelanguage.googleapis.com)

PROBLEM

Every generate_content request with input larger than ~500 tokens fails with HTTP 429

“Resource has been exhausted (e.g. check quota)”. Small prompts sometimes succeed (until a tiny

free-tier budget is used up), but any real workload (our chatbot sends ~16k input tokens per call)

fails 100% of the time — even at ~10 requests/minute.

The paid Tier 1 quota is clearly NOT being enforced; the project appears pinned to the

free-tier generate_content quota (which for gemini-2.5-flash is effectively 0 for real usage).

DASHBOARD (contradicts the 429s)

Rate Limit tab (Tier 1) shows huge headroom over the last day:

  • RPM: 15 / 1,000

  • TPM: 180.71K / 1,000,000

  • RPD: 207 / 10,000

Spend tab: ~đ129,322 this month, no spend cap set, total cost đ0 (covered by credit).

So neither rate limits nor spend cap are hit, yet every substantial request returns 429.

WHAT WE ALREADY TRIED (please don’t ask us to repeat)

1. Verified billing account is enabled, open, valid card, no balance due.

2. Confirmed the project is Tier 1 Postpay.

3. Created a BRAND NEW API key after billing was active — the new key ALSO returns 429

(so this is NOT a per-key free-tier pinning issue; it is project-wide).

4. Both our old and new keys on this project fail identically.

KEY EVIDENCE (this is a Gemini-API-specific enforcement issue)

Vertex AI (aiplatform.googleapis.com) on the SAME project restai-472508 works perfectly with

the same paid billing — gemini-2.5-flash serves normally there. Only the Gemini API /

AI Studio (generativelanguage.googleapis.com) returns 429. So GCP billing is valid; the problem

is specifically that the Gemini API paid Tier 1 is not being applied to this project.

REQUEST

Please investigate why project restai-472508’s Gemini API (generativelanguage) is enforcing

free-tier quota instead of the paid Tier 1 it is assigned to, and fix the tier binding so

generate_content requests are served under the paid Tier 1 limits (1M TPM / 1K RPM).

Region: requests from a server in Vietnam (asia). Happy to provide request IDs / timestamps.

1 Like