PuppyIP Resource Center
AI Tool Troubleshooting 5 min Published 2026-11-20

How to configure GPT-Live WebRTC standby and verify session charges

To pause GPT-Live inference during a quiet call, enable speech wake when you create a native WebRTC session and set an idle timeout of 5 to 60 seconds. Confirm standby and wake with session.paused and session.resumed.

GPT-Live WebRTC Voice API Standby billing

Service eligibility and regional restrictions

PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.

Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.

Key Takeaways

  • Native WebRTC standby works with GPT-Live-1 and GPT-Live-1.1. Configure speech wake and an idle timeout, and keep the microphone track active.
  • Use session.paused and session.resumed to verify the state. Time spent waiting for the idle timeout still counts as active.
  • Voice session duration is metered per second at $0.05 per minute active and $0.02 per minute standby. Backend models and tools are billed separately.

Configure speech wake and idle timeout at session creation

OpenAI's November 19 API changelog lists GPT-Live-1.1 and WebRTC standby. The standby setting applies to both GPT-Live-1 and GPT-Live-1.1. This guide covers an app using OpenAI's native WebRTC connection. For another gateway's WebSocket transport and billing, check that gateway's documentation.

At session creation, set <code>session.wake_policy.type</code> to <code>speech</code> and <code>session.idle_timeout_s</code> to an integer from 5 through 60. With 15 seconds, for example, speech wakes inference the first time, and 15 seconds without an input or output transcript can trigger standby. If you omit the timeout, inference stays active after the first wake until the session ends.

Verify standby and wake using session events

In a test session, subscribe to <code>session.resumed</code> and <code>session.paused</code> on the sideband connection. You can also listen on a WebRTC data channel configured to receive those events. Say a test phrase and note the first <code>session.resumed</code>. Stop speaking, wait for the last output transcript, and check for <code>session.paused</code>.

Keep the user-authorized microphone track active during standby. Speak again and verify another <code>session.resumed</code> with the same session ID. OpenAI says the conversation history carries through, and a short utterance may get a slower first response after waking. If you hear silence but receive no pause event, check the timeout at creation, whether output transcripts are still arriving, and whether the listener receives these events.

Separate active, standby, and backend usage

The pricing table lists the same voice session rates for both models, $0.05 per minute active and $0.02 per minute standby, metered per second. Silence while the timeout is counting down is active time. Waiting on a backend while inference continues is active as well. Muting the microphone alone does not pause active inference.

Estimate using the observed durations. Sixty active seconds and thirty standby seconds cost about $0.06 for voice. Backend model and tool charges come separately. Default WebRTC session creation bills 15 seconds at initialization and credits that duration once the session starts running. Count it once in the duration estimate for a running session.

Capture the final usage when closing

Keep pause and resume timestamps as application-side evidence for the split between active and standby. <code>usage.seconds</code> is a cumulative voice duration, so use the latest value. Listen for <code>session.closed</code> before initiating close and record the final seconds when it arrives. Track backend model usage separately and reconcile each with billing.

Sources

Frequently Asked Questions

Can I set a 90-second idle timeout?

The native WebRTC standby timeout accepts an integer between 5 and 60 seconds.

Can I estimate at the standby rate without a session.paused event?

First verify the session state and event subscription. The active rate applies while inference is running.

Does waking from standby create a new session?

The connection and session ID remain open during standby. Speech wakes inference and restores the conversation history.