OpenAI The most intelligent voice model: GPT-Live-1/mini comes on the scene, listening and speaking at the same time makes AI dialogue closer to real-person chatting

IT House reported on July 9 that OpenAI released the GPT-Live series of models today (July 9), positioning a new generation of full-duplex voice models that can listen and speak at the same time, making AI conversations feel closer to real people. The GPT-Live series models are currently available in two versions: GPT-Live-1 and GPT-Live-1 mini, which will be open to ChatGPT users around the world from now on. The GPT-Live-1 model is officially open to Go, Plus and Pro subscribers, and free users can call the GPT-Live-1 mini model. IT Home has attached relevant videos as follows: In terms of architecture, the GPT-Live series models adopt a full-duplex architecture, which no longer relies on responding after a single round. It can continuously process input and output at the same time. The model can judge whether to speak, continue listening, pause, interrupt or call tools "multiple times per second", thus making the conversation closer to the real-person chat experience. When faced with complex tasks, the GPT-Live series models split processing of continuous interaction and deep reasoning. The model maintains the voice conversation flow in the foreground, while delegating web search, complex reasoning or more complex tasks to the GPT-5.5 series of models for background execution: GPT-Live-1 (instant) and GPT-Live-1 mini call GPT-5.5 Instant GPT-Live-1 Medium calls GPT-5.5 Thinking Medium Inference intensity GPT-Live-1 High uses GPT-5.5 In terms of Thinking High reasoning strength and performance, OpenAI introduced that the company has established a new manual evaluation system, focusing on the comfort (pleasantness) and overall communication fluency (flow of conversation) in voice communication. Compared with the previous Advanced Voice