OpenAI has released the GPT-Live-1 API, allowing developers to build applications where speech models listen and talk simultaneously. This full-duplex capability, currently active within ChatGPT, enables pairing with various backend models to match reasoning depth, speed, and cost to specific tasks. Yelp is already using the system for phone-based reservations and reports improved call handling according to CTO Alex Levy.
The model performs significantly better on OpenAI’s benchmarks compared to earlier versions. In full-duplex interactivity tests, it scores 80.1 percent compared to 45.4 percent for GPT-Realtime-2.1. Turn-taking latency drops to 0.8 seconds from 1.4 seconds while tool-calling accuracy jumps to 87 percent from 60 percent. Banking voice support benchmarks show a 32 percent pass rate, up from 12.4 percent for the previous model. The release includes twelve new voices spanning different accents, dialects, and languages with ASR transcripts and response text provided out of the box. The service costs $0.05 per minute.
* Pricing is set at $0.05 per minute
* Turn-taking latency is now 0.8 seconds
* Banking voice support pass rate reached 32 percent
