Hello everyone, we're using LLM_MODEL="gemini-live...
# agents
s
Hello everyone, we're using LLM_MODEL="gemini-live-2.5-flash-preview" and recently received an email from the Gemini API Team stating that they will be retiring the following models on 18th November and we are advised to upgrade to newer ones. The following model endpoints will be removed on November 18, 2025. Model list • models/gemini-2.0-flash-lite-preview • models/gemini-2.0-flash-lite-preview-02-05 • models/gemini-2.0-flash-thinking-exp • models/gemini-2.0-flash-thinking-exp-01-21 • models/gemini-2.0-flash-thinking-exp-1219 • models/gemini-2.5-flash-lite-preview-06-17 • models/gemini-2.5-flash-preview-05-20 • models/gemini-2.5-pro-preview-06-05 • models/gemini-2.5-pro-preview-03-25 • models/gemini-2.5-pro-preview-05-06 What you need to do To avoid any service disruption, you must upgrade to one of the models listed below (or any upcoming models that we release) before November 18, 2025. These models offer improved performance and capabilities. • Gemini Pro models (
gemini-2.5-pro
) • Gemini Flash models (
gemini-2.5-flash
,
gemini-2.5-flash-preview-09-2025
) • Gemini Flash Lite models (
gemini-2.5-flash-lite
,
gemini-2.5-flash-lite-preview-09-2025
) I want to know which one would be the ideal one to switch to and if anyone here has worked on these.
c
what about TTS models?
s
we are using this realtime setup: session = AgentSession[UserData]( userdata=userdata, preemptive_generation=True, llm=google.beta.realtime.RealtimeModel( model=LLM_CONFIG["model"], voice=LLM_CONFIG["voice"], temperature=LLM_CONFIG["temperature"], ), ) I get that the name might seem misleading but this is not a separate stt, llm, tts pipeline, rather a realtime one
c
I want only the tts, is it possible?
h
current version available for gemini is : "gemini-2.5-flash-native-audio-preview-09-2025". as they are scheduled for deprecation: • gemini-2.0-flash-live-001 — Retiring on December 9, 2025 • gemini-live-2.5-flash-preview — Retiring on December 9, 2025
However, I have observed latency in this model. If you try it out and experience the same issue, please let me know.
s
i am using it right now. Apart from latency, another issue I am facing is that the agent sometimes gets stuck at a word, for exammple, it will say "Yessssssssss..." and the sss sound continues. I asked the agent why it did that and it said that it was waiting for me to finish speaking. But I was not speaking and even if it is waiting for me to finish speaking, it should be quiet and not producing the "sss" sound
another concern is that preview models can't be used for production, they are experimental features. Eventually, they get deprecated. Do you know of any stable realtime models suitable for production.