OpenAI opens GPT-Live-1 speech API at $0.05 per minute

GPT-Live-1 is not a brand-new model launch so much as a packaging decision: the speech model is already running inside ChatGPT, and OpenAI is now exposing it through an API so developers can build their own voice applications. The company is letting developers pair the speech layer with different backend models depending on the task, matching reasoning depth, speed, and cost to each use case.

The benchmark story is the pitch. Against GPT-Realtime-2.1, OpenAI reports full-duplex interactivity jumping to 80.1 percent from 45.4 percent, turn-taking latency dropping to 0.8 seconds from 1.4 seconds, and tool-calling accuracy rising to 87 percent from 60 percent. In a banking voice support benchmark, GPT-Live-1 reaches a 32 percent pass rate, up from 12.4 percent. These are OpenAI's own numbers, so the test is whether outside developers see the same gains.

The early customer example is Yelp, which is using the model for phone-based reservations and, according to CTO Alex Levy, reports better call handling. For companies weighing voice agents, the outcome likely hinges on whether that kind of operational improvement justifies $0.05 per minute at call volumes, and on whether the twelve new voices and built-in ASR transcripts and response text make integration straightforward enough to adopt.