OpenAI has released GPT-Live-1 in its API for developers building voice applications. The model listens and speaks at the same time rather than taking strict turns. It handles interruptions and short acknowledgements as they happen. Developers set tone, pace and style through the system prompt. They can also choose their own reasoning model and agent framework behind it. Telephony uses are aimed at reservations, order updates and customer service. Pricing is per minute, with backend model and agent fees billed separately. Early users include Yelp Host, Speak, Intercom's Fin and Cognition's Devin. One customer said it cut its voice codebase sharply. OpenAI says more voices and languages will follow, and points larger firms to Presence.
What changed
Voice agents chained separate speech recognition, reasoning and speech synthesis steps.
What it unlocks
Building phone and app voice agents that listen and speak at the same time and handle interruptions.
- $0.05 per minute in the API
- 12 voices at launch
- 80% fewer interruptions (Speak)
- +30 points on Full Duplex Bench
What you need to act on it
- OpenAI API access
- separate backend model and agent harness fees
- contact sales for custom voices
Sources