OpenAI 新型 GPT-Live-1 消除延遲——語音終能取代螢幕？
openai 推出了低延遲的全雙工模型 gpt-live-1 和 gpt-live-1 mini，取代了舊版語音模式。這表明語音正成為複雜、agentic ai 工作流的主要介面。
openai released gpt-live-1 and gpt-live-1 mini, full-duplex conversational models designed to handle natural turn-taking and real-time translation by processing speech and audio simultaneously. the system replaces the legacy cascade of speech-to-text, llm, and text-to-speech models, routing queries to advanced text models like gpt-5.5 for reasoning and context retention. currently rolling out to paid chatgpt tiers, the technology supports conversations lasting over 30 minutes and aims to enable 'hands-free' complex work. this development positions voice as a dominant interface for agents, competing with similar conversational updates from apple, amazon, and startups like sesame.