GPT-Live-1 边听边说，但延迟可能还是会让体验打折扣
openai 发布了 gpt-live-1，这是一个支持同时处理音频的全双工语音模型，承诺为付费订阅者带来更像人类的对话体验。
openai released gpt-live-1, a 'full duplex' voice model designed to handle concurrent input and output streams, eliminating the rigid turn-taking of previous versions. the model routes complex queries to gpt-5.5 for better reasoning and introduces real-time translation and visual data overlays for topics like finance and weather. available now for paid subscribers (plus gpt-live-mini-1 for free users), the update positions openai to compete with dedicated ai hardware and voice agents, while addressing safety concerns with built-in crisis support and anti-harm safeguards.