OpenAI has introduced GPT-Live, a new generation of voice models. This technology is already integrated into the "Conversation" mode of the ChatGPT chatbot.
The standout feature of GPT-Live is its full-duplex architecture. The neural network can now listen and speak simultaneously without waiting for silence. The model understands interjections, can pause, and does not interrupt users when they are simply thinking.
Developers have separated the communication process from task execution. While GPT-Live engages in dialogue, another model (currently GPT-5.5) runs in the background to search the internet for information or perform complex calculations. This allows the AI to translate speech in real-time and address queries without interrupting the conversation.
Source: OpenAI.Test results have shown that GPT-Live outperforms previous solutions in scientific reasoning and data retrieval. The ChatGPT interface has also introduced visual cards displaying weather, stock prices, and sports results.
OpenAI has released two versions: the powerful GPT-Live-1 and the simplified GPT-Live-1 mini. Availability:
- Plus and Pro subscription users have access to the main model;
- The mini version is available in the free ChatGPT version;
- The company will soon open an API for developers.
Programmers have added new safety filters to the system. These block attempts to mimic others' voices and restrict content related to violence or self-harm. Parental controls are provided for teenagers.
Recall that in June, OpenAI opened limited access to GPT-5.6 Sol, Terra, and Luna via API and Codex for a small group of trusted partners. The company launched this preview at the request of U.S. authorities.