Overview
GPT-Live-1 is OpenAI's full-duplex voice model, reachable in the OpenAI API as the model id `gpt-live-1` from September 10, 2026. It takes audio or text as input and returns audio or text as output. Image and video are not supported.
Full duplex is the defining property: the model keeps listening while it is still speaking, so a caller can interrupt mid-sentence and the reply adapts instead of finishing its turn. OpenAI's model card describes this as smooth interruption handling, and pairs it with streaming and function calling so a live conversation can trigger real work while it runs.
The voice layer is deliberately thin. OpenAI's documentation states that GPT-Live-1 delegates deeper reasoning and actions to the backend models and tools it is paired with, which is why its own knowledge cutoff of July 31, 2025 matters less than the model sitting behind it. Voice sessions are priced at $0.05 per minute, billed per second, with backend model and tool usage charged separately.
| Released | 2026-09-10 |
|---|---|
| License | Proprietary |
| Weights | API only |
| Knowledge cutoff | July 31, 2025 |
| Modalities | Audio, Text |
| Status | Generally available |
Strengths
- Full-duplex conversation — listens and speaks simultaneously, with smooth interruption handling
- Streaming and function calling, so a live call can trigger lookups or actions without ending the turn
- Delegates reasoning and actions to paired backend models and tools rather than doing that work in the voice layer
- Per-minute billing at $0.05, charged by the second, which makes the cost of a call easy to model
Best for
- Reach for it for phone-style assistants and support lines where a caller needs to be able to cut in.
- Reach for it for tutoring and practice-conversation products that break down when the learner has to wait for a full reply.
- Reach for it as the voice front end to an existing agent stack, with a stronger model doing the reasoning behind it.
How to access
| Provider | Model ID |
|---|---|
| OpenAI API ↗ | gpt-live-1 |
FAQ
How much does GPT-Live-1 cost?
GPT-Live-1 voice sessions cost $0.05 per minute and are billed by the second. Backend model usage and tool calls made during the session are charged separately at their normal API rates, so the full cost of a conversation depends on how much work GPT-Live-1 delegates to the model and tools behind it.
Is GPT-Live-1 available on OpenAI's free tier?
No. GPT-Live-1 is unsupported on the free tier. Concurrency scales with paid usage tier instead: Tier 1 accounts get 25 concurrent voice sessions and Tier 5 accounts get 500, so how many callers an application can handle at once depends on its spending tier rather than a separate quota.
What does full duplex mean for GPT-Live-1?
Full duplex means GPT-Live-1 listens and speaks at the same time rather than taking strict turns. A caller can interrupt while the model is mid-sentence, and OpenAI's model card describes the result as smooth interruption handling. Turn-taking voice models instead force the caller to wait for a complete reply before speaking again.
Can GPT-Live-1 do complex reasoning on its own?
GPT-Live-1 delegates deeper reasoning and actions to the backend models and tools it is paired with, according to OpenAI's documentation. The model supports streaming and function calling to route that work outward during a live call. Its own knowledge cutoff is July 31, 2025, so current information has to come from the tools or model behind it.