{"reason": "error", "error": "type='realtime_model_error' timestamp=1765839019.5552053 label='livekit.plugins.google.realtime.realtime_api.RealtimeModel' error=ConnectionClosedError(Close(code=1007, reason='Failed to select an endpoint for the request: The request is not supported by this model.'), Close(code=1007, reason='Failed to select an endpoint for the request: The request is not supported by this model.'), True) recoverable=False"}

Thanks for reaching out, Phanos. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

gemini-live-2.5-flash-native-audio in vertex Other models work ok in vertex. The same model works well in google ai studio

Thanks — I’m responding about the error when calling the gemini-live-2.5-flash-native-audio model in Vertex where the request fails with "The request is not supported by this model," while the same model works in AI Studio and other models work in Vertex. Likely causes and what you can check • The model isn’t enabled for the specific realtime/streaming feature in Vertex. AI Studio can expose capabilities that aren’t available or routed the same way from Vertex. • The Vertex endpoint/region or selection logic may be routing to an endpoint that doesn’t support the “native audio” realtime variant. • Your Vertex request includes parameters or features (streaming audio input, response_format, or plugin-specific flags) that this model’s Vertex integration doesn’t accept. • IAM/quotas or model provisioning differences between Studio and Vertex can produce rejections that surface as "unsupported" errors. Concrete troubleshooting steps 1. Confirm model support in Vertex ◦ Verify that gemini-live-2.5-flash-native-audio is listed and supported for realtime/streaming use in your Vertex project/region. 1. Check the Vertex endpoint/region and model name ◦ Ensure the request targets the correct Vertex endpoint and region where the model is available. ◦ Use the exact model identifier that Vertex expects (no Studio-only aliases). 1. Compare request payloads ◦ Compare the working AI Studio request and your Vertex request. Look for differences in: ▪︎ streaming vs non-streaming flags ▪︎ audio input format fields ▪︎

response_format

or any plugin-specific parameters ▪︎ headers or content-type ◦ Share a redacted request snippet if you’d like me to spot incompatible fields. 1. Simplify the request ◦ Try a minimal request (no special realtime/audio features) to confirm basic model invocation succeeds in Vertex. If that works, add features back one at a time to find the unsupported parameter. 1. Inspect logs and error details ◦ Check Vertex request logs and any error detail fields. The server-side log may show which parameter triggered the rejection. 1. Region/endpoint fallback ◦ If possible, try calling the same model from another Vertex region that lists gemini-live variants to see if routing is the issue. If you’d like, paste a redacted copy of the Vertex request you’re sending (remove keys/secrets) and I’ll review it for incompatible parameters and suggest the minimal change to make it match the Studio call.