Back to all discussions

  •   3 months ago

Cannot get live Gemini models

I am totally new to Google Cloud and Live API. I followed all the guides in the resources and easily created agents using ADK, Google AI Studio and Vertex AI. But, when I tried using gemini-2.5-flash-native-audio-preview-12-2025 for building live agents, I found that it wasn't available on my API. I tried using Google AI Studio as well as Vertex AI. Still didn't work. I used the region as us-central1 and it still didn't work. I used a python script to check all available models and found extra models like gemma, computer-use, etc. through Vertex but still didn't find any model that supports BiDi streaming. How do I create a live agent without a live model?

  • 2 comments

  •   •   3 months ago

    Hey! I ran into the same confusion early on. A few things that helped:

    1. The Live API models won't show up when you list models through the standard Vertex AI or GenAI SDK model listing endpoints. They're only accessible through the Live API client specifically.

    2. The model name format matters. For native audio with bidi streaming, I'm using `gemini-live-2.5-flash-native-audio` (not the preview-12-2025 variant). Try that exact string.

    3. You need to use the `google-genai` SDK's Live API, not the standard `generateContent` path. Here's the basic pattern:

    ```python
    from google import genai
    from google.genai import types

    client = genai.Client(api_key="YOUR_KEY")

    config = types.LiveConnectConfig(
    response_modalities=["AUDIO"],
    speech_config=types.SpeechConfig(
    voice_config=types.VoiceConfig(
    prebuilt_voice_config=types.PrebuiltVoiceConfig(voice_name="Kore")
    )
    )
    )

    async with client.aio.live.connect(
    model="gemini-live-2.5-flash-native-audio",
    config=config
    ) as session:
    await session.send(input="Hello", end_of_turn=True)
    async for response in session.receive():
    # handle audio chunks
    pass
    ```

    4. Make sure you're using the latest `google-genai` SDK (`pip install -U google-genai`). Older versions don't have the Live API client.

    The key insight is that Live API is a completely separate connection path (bidirectional streaming over WebSocket) it's not an extension of the regular generate endpoint, which is why those models don't appear in standard model listings.

    Hope that helps!

  •   •   3 months ago

    Hi Shoeb,
    Try to use "gemini-2.5-flash-native-audio-preview-12-2025" it is worked for me with Google AI Studio

Log in or sign up for Devpost to join the conversation.