Gemini Live issues

Agents
1

I am trying to use gemini realtime model (gemini-live-2.5-flash-native-audio) for my voice agent and I am seeing following issues -

  1. Hallucinations

  2. Sometimes, model does not respond. You need to say hello 3-4 times before it generates any response.

  3. Could not pronounce some words properly.

If anyone is using gemini realtime model in production, can you suggest how are you ensuring that these issues are not occurring. (edited)

2

Have you checked the Gemini Live forums?

3

Found the same and also unreliable tool calling / repeating the response after each tool call .. concluded that Gemini 2.5 Flash Preview is faaaar from production-ready.

3 Likes
4

Did you use Google plugin from livekit?

5

Yeah I am using that only.