Hi everyone,
I’m running a low-volume agent (built with Google’s Agent Development Kit) that calls gemini-2.5-flash via Vertex AI a handful of times per execution (about 5-15 generateContent calls, with delays in between). My project is on a paid billing account, confirmed to be at Tier 1. Despite this, I keep getting 429 RESOURCE_EXHAUSTED errors.
Model: gemini-2.5-flash
Requests are routed to locations/global rather than a specific region.
Error:
429 RESOURCE_EXHAUSTED. {“error”: {“code”: 429, “message”: “Resource exhausted. Please try again later.”, “status”: “RESOURCE_EXHAUSTED”}}
This has happened even on the very first LLM call of a session, and even after waiting several minutes between attempts. I also waited a full 24 hours and tried again, expecting the issue to be a temporary propagation delay after the billing upgrade - but I got the exact same 429 error again, which makes me think it’s not just a transient issue.
I checked the “Request limit per model per minute for a project in the paid tier 1” quota for the Gemini API and it shows 1000 (which I understand is the maximum settable without going through sales), but the actual usage shown is 0%. So the quota itself doesn’t seem to be the bottleneck, at least not the one I’m looking at.
I don’t currently have a paid Cloud Support plan, so I can’t open a technical support case - that’s why I’m asking here.
Questions:
- If the “Request limit per model per minute” quota shows 0% usage, is there another quota or limit I should be checking instead (e.g., tokens-per-minute, concurrent requests, or something specific to the “locations/global” routing)?
- Are there any other ways to resolve or work around this 429 issue besides requesting a quota increase through sales?
Any insight into what’s actually causing this would be really appreciated. Thanks!