There’s another thread describing a similar issue - which appears to come from the same root cause as what is observed here.
In our case, we’re not using the chat interface but the API directly via the google-genai Python SDK, running into rate limit errors very quickly.