Persistent 429/403 lockout on Tier 1 Paid project despite usage at <4% of all limits — May 5-6

Gemini API
1

Hi team,

I’ve been locked out of Gemini 3.1 Pro Preview since approximately 06:00 UTC on May 5, 2026, with persistent 429 errors that do not match my actual usage. Spend cap is set to €100. Billing is confirmed active. The behavior is inconsistent with my dashboard data and I would appreciate AI Studio team review.

Project setup:

  • Tier: Paid Tier 1 (confirmed in Billing dashboard)

  • Models affected: Gemini 3.1 Pro Preview primarily; partial behavior on other models

The discrepancy:

My Rate Limit dashboard shows for the last 24 hours on Gemini 3.1 Pro:

  • RPM peak: 1 of 25 (4%)

  • TPM peak: 417.63K of 2M (21%)

  • RPD peak: 8 of 250 (3%)

For the last 7 days:

  • RPM peak: 3 of 25 (12%)

  • TPM peak: 552.92K of 2M (28%)

  • RPD peak: 92 of 250 (37%)

Spending: €35.21 of €100 cap for the current month. €93.63 over the last 28 days. No cap breach.

Despite this, the API returns 429 RESOURCE_EXHAUSTED on the majority of requests, with a smaller number of 403 Forbidden errors mixed in. Success rate dropped to 0% during a spike around 04:00 UTC on May 5, recovered briefly, then dropped again at approximately 06:00 UTC and has remained at near-zero throughput since. As of this post (01:14 UTC May 6), I can typically send one request and receive one response per session before the next request fails.

What I have already tried:

  • Waited 24+ hours past the spike for automatic recalibration. No change.

  • Raised spend cap from default to €100 per Google Cloud Billing support advice. No change.

  • Re-confirmed billing setup is correct. Billing support confirmed the account is in good standing and could not identify the cause. They advised escalating here.

  • Tested across multiple models. Behavior persists.

Request:

Could the AI Studio team please review my affected project for any flags, throttles, or tier-state inconsistencies that would not be visible in the user-facing dashboards? The pattern of 0% success rate at usage levels well below all documented quotas, combined with mixed 403 errors, suggests an issue beyond standard rate limiting.

Project ID, screenshots of usage dashboards, rate limit pages, error charts, and billing page available privately on request.

Thanks.