Regarding the Gemini Realtime Live API, the documentation states that the rate limit is up to 3 concurrent sessions per API key.
However, in my actual experiments, I created 3 separate API keys and tried to establish 5 concurrent sessions simultaneously using these keys, but I encountered an error on the 5th session.
While I understand that this feature is experimental and some discrepancies might occur, this outcome suggests that the documentation might be incorrect regarding the limit being strictly per API key.
Does this imply that the current effective limit is actually 3 concurrent sessions per user (or perhaps per project/account), rather than what the documentation currently states?
On a final note, I’ve found that even with the current performance and limitations, the API is already proving quite useful for my service. Therefore, I would greatly appreciate it if this concurrent session limit could be relaxed in the near future.