Date: 2026-09-28. Intended usage: unpaid Gemini API pilot; no billing upgrade requested.
We receive HTTP 503 with Google RPC status UNAVAILABLE from:
POST https://generativelanguage.googleapis.com/v1beta/models/gemini-3.5-flash-lite:generateContent
Three separate, manually initiated one-shot requests failed: two with one second of synthetic silence, and one with the following minimal text-only body:
{
"contents": [{"role": "user", "parts": [{"text": "Reply with exactly OK."}]}],
"generationConfig": {"maxOutputTokens": 64}
}
Requests used native Node.js fetch, Content-Type application/json and the existing API key in the x-goog-api-key header. No automatic retries, tools, dictionary, system instruction or thinking override were used in the text-only test. The text test deadline was 30 seconds; it returned HTTP503, not a local timeout.
Read-only models.list using the same server-side key returned HTTP200 and listed gemini-3.5-flash-lite with generateContent support. No additional pages were indicated.
AI Studio showed these 28-day peak/limit values for the same project and model:
- RPM: 1 / 15
- TPM: 110 / 250000
- RPD: 2 / 500
These are displayed historical peaks, not proof of current capacity or successful inference. The full provider error text and response headers were not retained; only HTTP and RPC status were recorded. No global incident has been confirmed.
Could you advise how to distinguish model capacity issues from project-specific inference availability, and what non-secret diagnostic metadata would help? We do not want to enable billing or rotate keys speculatively.
No credentials, private audio, account email, server IP or project identifiers are included in this report. Private identifiers can be provided only through an appropriate verified support channel if required.