Gemini Live API sessions hang when a query should trigger a function call

The problem

We are experiencing a critical issue with the Gemini Live API using the gemini-2.0-flash-live-001 model, specifically when attempting to use function calling. Starting approximately on Tuesday, May 13, 2025, around 4:00 AM UTC+3, API sessions began to hang indefinitely when a query is made that should trigger a function call. Symptoms: API Hangs: When a prompt intended to trigger a declared function is sent, the API session stops responding.

The fix

Gemini Live API sessions that used to work suddenly hanging on function calls, with the socket staying open, was a Google server-side incident on gemini-2.0-flash-live-001, fixed about two days after the report with no client change needed. Reusable lesson: reproduce the same call in AI Studio; if it hangs there too, it's a service-side incident, not your code, so file it and wait. Client-side, add reconnect with backoff on hung sessions so your app recovers on its own once the backend is fixed.

In gh:googleapis/python-genai#803 (https://github.com/googleapis/python-genai/issues/803), the reporter ran into this: We are experiencing a critical issue with the Gemini Live API using the gemini-2.0-flash-live-001 model, specifically when attempting to use function calling. Starting approximately on Tuesday, May 13, 2025, around 4:00 AM UTC+3, API sessions began to hang indefinitely when a query is made that should trigger a function call. Symptoms: API Hangs: When a prompt intended to trigger a declared function is sent, the API session stops responding. The verified resolution: Gemini Live API sessions that used to work suddenly hanging on function calls, with the socket staying open, was a Google server-side incident on gemini-2.0-flash-live-001, fixed about two days after the report with no client change needed. Reusable lesson: reproduce the same call in AI Studio; if it hangs there too, it's a service-side incident, not your code, so file it and wait. Client-side, add reconnect with backoff on hung sessions so your app recovers on its own once the backend is fixed.