I've seen a number of similar issues here with the same response code, but none of them match our experience.
- The server is located in Finland (allowed region in Gemini API)
- Requests are made from one IP V4
- GCP service account is used,
- SA has an API key with Generative API permissions
- SA has roles and permissions for Gemini and Vertex AI management
- We did not exceed rate limits
- We did not make any changes whatsoever to our code or GCP setup during the period when this started.
- Model used - Gemini 2.5-experimental (not preview)
- Paid tier level 1
Our issue started on April 17th, where all requests started to be blocked from our server in Finland. Local environment requests work fine.
Does anyone at all have any idea how to fix this? This severely disrupted our infra and SLA as we had to urgently migrate to OpenAI's models. This started out of the blue and there is no explanation I could find. Support for GCP is also unavailable, I have only found this single forum.
LinkedIn
Twitter
