GPT-Live-1 is now in the API

By OpenAI

Categories: AI, Product

Summary

OpenAI launched GPT-Live-1, a full-duplex conversational AI model priced at 5 cents per minute that handles real-time interruptions and background noise while delegating reasoning to backend models. This enables production-ready voice AI at scale with separate pricing for backend inference and tool services.

Key Takeaways

  1. Full-duplex architecture allows simultaneous listening and speaking, eliminating latency delays in voice conversations and enabling natural interruption handling.
  2. Pricing model separates front-end ($0.05/min) from backend inference and tools, allowing builders to optimize costs based on conversation complexity and reasoning needs.
  3. Two-tier model design pairs lightweight front-end for real-time conversation with backend model for reasoning and tool delegation, maintaining conversation flow while handling complex actions.
  4. Robust noise handling enables production deployment in real-world environments with background noise, expanding viable use cases beyond controlled settings.
  5. Direct API access positions this as drop-in replacement for ChatGPT Voice, enabling developers to integrate conversational AI into existing applications immediately.

Related topics

Transcript Excerpt

Hey, what are we launching today? We're launching GPT-Live 1 in the API. It's a full-duplex model for natural, expressive conversation, built to handle background noise and interruptions smoothly, while pairing with a back-end model for tools and reasoning. Think of it like ChatGPT Voice for all- Wait, you were saying I don't have to worry about interruptions anymore? Right, you can jump in. I can listen and speak at the same time. Sorry, there's some background noise here. Can you still hear me OK? Yeah, you're coming through clearly. I can follow your voice even with noise around. And you were saying something about a back-end model? Mm-hmm. GPT-Live 1 pairs with one for reasoning and tools. That's how I delegate actions to the robot and display here, while keeping the conversation flowi…

More from OpenAI