GPT-Live-1 Is in the API at $0.05 a Minute, but That Buys Only the Voice Layer

Verdict: GPT-Live-1 stays conditional. The condition we attached in July, no API, is gone. The one that replaces it is cost: the $0.05 per minute OpenAI charges covers only the voice session. The reasoning runs on a backend model you choose, such as GPT-5.6 Luna or GPT-6 Astra, billed separately, and OpenAI gives no all-in figure.

What happened

OpenAI's API changelog entry of 10 September and its developer docs say:

• Availability. "GPT-Live 1 is now generally available in the API."
• Price. "$0.05 per minute, billed per second". "Backend model and tool usage is billed separately."
• Backend choice. The backend can be "a supported Responses model" or "any model, agent harness, or service your application runs".
• Telephony. The getting-started guide lists "Telephony and SIP" for connecting phone calls.

We have not tested the API.

Why it matters

$0.05 a minute is $3 an hour of voice-session time. Every minute of that conversation also runs a backend model at its own token rate. Two OpenAI models from our directory as examples:

OpenAI's docs give no worked per-minute figure for any pairing, so the $0.05 number alone cannot budget a voice agent. The gap between those two backends is 50x on input tokens.

Our directory also lists GPT-6 Luna, released 22 September, at $0.10 / $0.50 per 1M tokens. We have not confirmed which models count as a "supported Responses model" for GPT-Live-1; check before you budget on any of them.

What changes for you

• Cost both halves. Run a representative call transcript through the voice session and your chosen backend, and compute the all-in cost per minute before you commit.
• Route by difficulty. Put high-volume, simple flows such as scheduling on a budget backend, and reserve a frontier model such as Astra for calls that need it.
• Confirm the backend. Check that your chosen model is supported before you budget on its rate.

Our directory. The GPT-Live-1 entry now carries the API pricing: $0.05 per minute for the voice session, with the backend model and tools billed separately. Its summary still describes the ChatGPT-only model, including "no API", and is flagged for refresh. The rating does not move: the no-API condition has cleared, but the all-in cost is unpublished and we have not tested the API.

FAQ

How much does GPT-Live-1 cost in the API? $0.05 per minute for the voice session, billed per second, per OpenAI. The backend model and tool usage are billed separately, so the real per-minute cost depends on which model you pair it with.

Can GPT-Live-1 answer phone calls? Yes. OpenAI's getting-started guide lists Telephony and SIP for connecting phone calls.

Which backend model should I use? OpenAI's docs allow a supported Responses model or any model, agent harness or service your application runs. Price your candidates against a real transcript before choosing.

Is GPT-Live-1 generally available? Yes. OpenAI's API changelog lists GPT-Live 1 as generally available in the API from 10 September.