Past few days, gemini-2.5-flash API throws 429 Too Many Requests error frequently. Dashboard shows token usage over TMP like below, but it does work without any error on last week with almost same input. So I think that there is some issue about token-counting in last few days. Also, when 429 error occurred, however, even though I stopped the application, the token limit per minute remained over the threshold for more than 10 minutes. So wait-and-retry solution does not help. What happen on Gemini API? Why per minute limit does not reset after over minute?
My app uses gemini-2.5-flash model with OpenAI compatible API, paid tier 1.
