OpenAI released GPT-Live-1 in the API on September 10.

It can listen and speak at the same time. OpenAI says it can also hand deeper reasoning and tool calls to a backend model.

The announced price is $0.05 per minute for the voice layer. That is not an all-in price for the backend work. This is an announcement summary, not a hands-on review.

A small prototype idea: a voice guide for a fictional app's help page.

Give it a short help document with just a few answers. Keep the first version focused on explaining things, with no account changes or purchases.

Then try conversations that aren't perfectly scripted:

• Start a question, pause to think, then finish it. Does the guide leave room for you?

• Interrupt its answer with "wait, I meant the mobile app." Does the next answer follow the correction?

• Ask something the help document doesn't cover. Does it say what it can't answer?

• Ask it to repeat one step slowly. Does it repeat the useful step without restarting the whole explanation?

These are suggested checks for a prototype, not results from testing this model.

For each attempt, write down what you said, what happened, and what you expected. Include a text version of the help so people can still read the instructions.

The interesting question for a builder is whether voice makes that particular task easier. A natural-sounding demo is a starting point; the awkward conversations help you decide what to improve.

Official announcement and pricing scope: https://openai.com/index/introducing-gpt-live-1-in-the-api/