GPT-Live-1.1 API adds lower-cost voice standby
OpenAI released GPT-Live-1.1 in its API on 19 November, with a new standby option for WebRTC voice sessions. The release gives commerce developers another option for building spoken support…
OpenAI released GPT-Live-1.1 in its API on 19 November, with a new standby option for WebRTC voice sessions. The release gives commerce developers another option for building spoken support workflows, including conversations that pause while a customer finds an order number or receipt.
The dated API changelog identifies improvements to pronunciation, accent consistency and multilingual transcription. This is a new API availability announcement; the model had already powered voice in ChatGPT Plus and Pro since October, according to OpenAI Developers.
What the minute prices include
The model page lists active voice sessions at $0.05 per minute and standby at $0.02 per minute. Backend models and tools are charged separately. The lower rate applies while inference is paused in a WebRTC session; it is not a blanket discount for every phone integration.
For illustration, eight active minutes and two standby minutes cost $0.44 in voice-session charges, compared with $0.50 for ten active minutes. That is our arithmetic using the published rates, not a measured support-call saving. It excludes backend usage and any separate telephony costs.
Configure the pause, then test the wake-up
OpenAI’s session guide describes a speech wake policy and an idle timeout from 5 to 60 seconds. The connection and session identity remain intact during standby, and conversation history is restored as speech resumes. OpenAI cautions that the first response to a short utterance after standby can take longer.
Our recommendation is to test pauses that resemble your customers’ actual calls: looking for an email, reading a tracking number, correcting an address or switching languages. Record latency as well as cost. A lower idle bill is less useful if callers repeatedly think the assistant has disconnected.
A voice interface still needs a reliable order workflow
GPT-Live-1.1 can listen while speaking and delegate reasoning or tool work to a backend. It does not supply a ready-made Shopify order-support integration. The app still has to authenticate the customer, retrieve the right order and decide which changes require confirmation.
Start a trial with read-only order status and compare spoken identifiers against the records actually retrieved. Include deliberately similar names and order numbers. Keep a clear route to a person when the system cannot identify the order or explain a discrepancy. The Founder’s Inn has not benchmarked the model’s accuracy on merchant calls.
More from AI
See all →
