OpenAI GPT-Live-1 Lets AI Listen and Speak at the Same Time

OpenAI opened developer access to its voice model GPT-Live-1 via API: it listens and speaks simultaneously, supports interruptions and topic changes. Response latency is 0.798 seconds in Full Duplex Bench, tool-calling accuracy is 87%. Price is $0.05 per minute.

• Model works in full duplex: listens while speaking
• Response latency 0.798 s in Full Duplex Bench
• Tool-calling accuracy is 87%
• API access costs $0.05 per minute