AI & Machine Learning

OpenAI Opens GPT-Live-1.1 API With Cheaper Voice Standby

Developers can now use ChatGPT’s GPT-Live-1.1 voice model, with improved multilingual speech and a lower rate when calls are on standby.

By Daniel Mercer Edited by Maria Konash Published: Updated:
OpenAI Opens GPT-Live-1.1 API With Cheaper Voice Standby
OpenAI extends GPT-Live-1.1 to developers and introduces lower voice duration charges during standby. Photo: OpenAI 

Key Notes

  • OpenAI has released GPT-Live-1.1 in the API after introducing the model in ChatGPT Plus and Pro in October, with claimed improvements in multilingual speech and transcription.
  • WebRTC sessions can enter standby after a configurable idle period, reducing the voice duration rate from five cents to two cents per minute during those pauses.
  • The voice model can keep a conversation going while a separate backend performs work, with backend model and tool usage billed separately.

OpenAI has opened GPT-Live-1.1 to developers, bringing the voice model already used in ChatGPT to applications built through its API. The release targets assistants that can hold a conversation while an AI agent looks up information or completes work in the background.

The November 19 API changelog announces improvements in multilingual pronunciation, accent consistency and transcription. OpenAI also introduced a cheaper standby setting for quiet stretches of a call, addressing the cost of keeping a voice assistant available between requests.

ChatGPT’s Voice Model Moves Into Other Apps

GPT-Live-1.1 can listen and speak simultaneously, allowing a person to interrupt, add a detail or change direction before the assistant finishes its reply. OpenAI says it handles precise information such as letters and numbers more reliably, an important distinction for conversations involving names, addresses or account references.

The model has powered voice in ChatGPT Plus and Pro since October 15, according to the developer announcement. The new release extends that capability to developers using gpt-live-1.1. It builds on the full-duplex approach we covered when the GPT-Live family launched in July.

OpenAI’s model documentation lists audio and text as supported inputs and outputs. The API model does not accept images or video, so its availability should not be read as a release of every capability found elsewhere in the ChatGPT app.

Standby Cuts the Cost of Quiet Time

GPT-Live-1.1 costs $0.05 per active minute and $0.02 per standby minute, with voice sessions billed by the second. That makes the standby duration rate 60% lower than the active rate. Backend model and tool charges remain separate, according to OpenAI’s pricing.

Standby is available for both GPT-Live-1 and GPT-Live-1.1 over WebRTC. Developers can set an idle timeout between five and 60 seconds, after which inference pauses while the connection stays open. The conversation resumes when the user speaks again.

The session guide notes that conversation history is restored on resumption, although the first response to a short utterance may take longer. Developers must configure the timeout: if it is omitted, inference remains active after the first wake until the session ends.

Conversation and Execution Have Separate Jobs

GPT-Live handles the spoken exchange and delegates reasoning or tool use to a backend. That backend can be a configured OpenAI Responses model or a developer’s own model, agent or service. An illustrative customer-service workflow could keep talking to a caller while a separate system checks an order and returns its status.

The delegation documentation assigns permissions, required confirmations and business records to the application. Developers therefore still decide which operations an assistant may perform and when a user must approve them. A more natural voice does not establish that an underlying transaction was authorized or completed.

For businesses, the combination is more consequential than an accent upgrade alone. A voice interface that remains responsive during background work can make an agent easier to use, while standby reduces the duration charge during pauses. The practical measure is whether the complete service understands the request, finishes the task correctly and keeps the total cost acceptable.

Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.

AI & Machine Learning, Enterprise Tech, News