Newshot

OpenAI opens GPT-Live-1.1 API: supports real-time two-way conversation, agents can chat while handling complex tasks

AI summary

OpenAI announced the opening of the GPT-Live-1.1 API voice model to developers, allowing third-party developers to integrate it into their applications to build AI voice agents that can listen, respond, and execute complex tasks in real time. GPT-Live-1.1, an upgraded version of GPT-Live-1, features full-duplex voice interaction, can recognize pauses, interruptions, and speech rate changes, and supports delegating tasks requiring deeper reasoning or tool calls to a backend model chosen by the developer while voice conversation continues uninterrupted.

Translation(model translation, not the source's own text)

OpenAI Opens GPT-Live-1.1 API: Real-Time Two-Way Conversation, Agents Can Chat While Handling Complex Tasks

IT之家 November 20 news: OpenAI today announced it is opening the GPT-Live-1.1 API voice model to developers, allowing third-party developers to integrate it into their own applications to build AI voice agents capable of listening, responding, and executing complex tasks in real time.

According to OpenAI, GPT-Live-1.1 is an upgraded version of GPT-Live-1, featuring full-duplex voice interaction capability: it can listen and speak simultaneously, without having to wait for the user to finish a full sentence before beginning processing. The model can recognize pauses, interruptions, and changes in speaking rate during a user's speech, and judge in real time whether it should keep listening or respond.

IT之家 note: Since October 15, 2026, GPT-Live-1.1 has been the default voice model for ChatGPT Plus and higher-tier subscribers. This API opening further extends the relevant capabilities to third-party developers.

In terms of specific capabilities, OpenAI says GPT-Live-1.1 improves multilingual voice processing, instruction following, and voice fidelity, and can more reliably handle information requiring precise recognition, such as letters and numbers, reducing comprehension deviations during voice interaction.

In addition, the model supports delegating tasks requiring deeper reasoning or tool calls to a backend model of the developer's choice. Developers can use either OpenAI's models or connect third-party models. While the backend model handles complex tasks, the voice conversation can continue without interrupting the current interaction.

This means developers can combine real-time voice interaction with the reasoning and tool-calling capabilities of backend models, enabling voice agents to handle complex multi-step workflows while answering user questions.

According to OpenAI's official API documentation, the GPT-Live series uses a full-duplex architecture that can directly process input and output audio, reducing the latency caused by chaining speech recognition (STT), large language models (LLM), and speech synthesis (TTS) in traditional voice systems.

OpenAI has provided technical documentation and integration guides for the GPT-Live-1.1 API; developers can learn about the model's capabilities, configuration methods, and specific usage through the OpenAI developer platform.

Related Articles

  • Apple is developing its own AI agent: able to control devices, browse the web, and write code, aiming to compete with OpenAI and Anthropic
  • ChatGPT adds "real-time dictation": speak into the microphone and speech turns into text instantly
  • Due to agents intruding into banking systems, Arizona launches criminal investigation into OpenAI and other AI companies
  • Report says OpenAI is entering the phone market; FIH Mobile under Foxconn wins assembly orders
  • Jensen Huang's departure sparks tech industry discussion: Musk says he is irreplaceable, Microsoft's Nadella advises against idolizing founders
  • Supply chain confirms OpenAI's first AI hardware to start shipping by year-end: co-created with former Apple designer Jony Ive, with Chinese manufacturers such as Luxshare Precision deeply involved

The translation is machine-generated and individual terms may be off. Nothing is rewritten or cut; the original is authoritative.

Published by IT之家
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗