I am using Google AI Studio Generate Speech with gemini-3.1-flash-tts-preview.
Issue
- The project is shown as Tier 1 / Prepay.
- The Prepay credit balance is positive.
- Cloud Billing shows a Paid account and recorded the cost of the first successful generation.
- The first TTS generation succeeded on August 9, 2026.
- Every subsequent generation returns:
Http response at 400 or 500 level, error: , http status code: 403 - The problem reproduces with a one-sentence Japanese prompt and default voice/settings.
- A newly created API key linked to the same project returns the same 403.
- Usage is within the displayed limits: RPM 1/10, TPM 355/10K, RPD 1/100.
- No warning or appeal banner appears in AI Studio or Google Cloud Console.
- The API keys were created in AI Studio and are restricted to the Gemini API by default.
Expected behavior
TTS generation should continue while the paid Tier 1 project has a positive Prepay balance and usage remains within limits.
Actual behavior
All Generate Speech attempts after the first successful request fail with HTTP 403.
Google Cloud Billing Support confirmed that this technical issue is outside their scope and directed me to report it here.
Could the Google team check whether the paid project/account has been restricted by mistake or whether there is a billing/access synchronization issue? I can provide the project ID privately to authorized Google staff, but I am not posting it publicly.