**[OpenAI Launches GPT-Live-1 Full-Duplex Voice Model in API]** According to a report by Jinse Finance, on September 10, OpenAI officially announced the launch of the GPT-Live-1 full-duplex voice model in its API. The motivation behind this development stems from the limitations of traditional voice agents, which require the integration of three components: speech recognition, large language models, and speech synthesis. Each handoff introduces latency and risks losing conversational flow and context. After extensive research and development, GPT-Live-1 integrates listening and speaking into a single model, enabling real-time interruption handling and confirmations, while delegating deep reasoning tasks to backend models such as GPT-6Astra or Luna. Results show that this model outperforms GPT-Realtime-2.1 by 30 percentage points in full-duplex benchmark tests. When paired with GPT-6Astra, it ranked first in the Tau3 voice intelligence evaluation. Early assessments by the language learning platform Speak indicate an 80% reduction in interruptions during thought pauses, while a healthcare client reported an 80% simplification of their codebase, reducing 23,000 lines of code. The model supports telephony deployment, 12 new voice options, and front-end voice layer pricing at $0.05 per minute. It is now available in the API, with companies such as Yelp, Fin, and Cognition already participating in testing.