Is anyone using gemini 2.5 flash in production for their voice agent? We started experimenting with it and conversations feel smoother since time to first token is quicker and seems more consistent that gpt 4o, and content seems about the same. I'm considering switching but I don't much chatter about using it in production for voice agents.
b
brave-island-45242
11/21/2025, 9:07 PM
A couple things to keep in mind:
• Might be worth it to use the google plugin since it has fixes for some of google's weird apis behaviours
• If you are going to use vertex be mindful of ratelimit because they use a shared dynamic quota
Gemini models have some oddities that you will have to handle. For example, empty responses. But all in all is not a bad model to use.
g
green-match-33925
11/21/2025, 10:32 PM
the gemini 2.5 flash model is great for voice ai application, however how to handle that intermittent empty responses ? any fix for that ?