OpenAI GPT Live 1 API Availability
GPT Live 1 is generally available in the OpenAI API for full-duplex voice sessions. Learn about pricing, delegation options, and evaluation steps.
General Availability Announcement
On September 10, 2026, OpenAI confirmed that GPT Live 1 is generally available in its API. The announcement specifies the use of the v1/live/sessions endpoint to facilitate full-duplex voice conversations. A key characteristic of this release is the ability for voice interactions to continue while a backend model or agent handles reasoning and tool execution simultaneously.
Delegation and Pricing Structure
The source documents two delegation methods for handling backend logic during live sessions. Developers can use Responses delegation with an OpenAI model or utilize client delegation to connect their own backend infrastructure. Regarding costs, voice sessions are priced at $0.05 per minute, with billing calculated per second. It is explicitly stated that backend model and tool usage are charged separately from the voice session fee.
- Voice session cost: $0.05 per minute (billed per second).
- Backend model and tool usage: Charged separately.
- Delegation options: Responses delegation (OpenAI model) or client delegation (custom backend).
Evaluation Procedure and Limitations
To evaluate GPT Live 1, engineers should design tests that measure latency and audio continuity during concurrent backend processing. A practical procedure involves initiating a live session, triggering a complex tool call, and measuring the interruption time in the voice stream. Compare these metrics against baseline non-live models to validate the full-duplex claim. However, the source does not provide benchmark data, context window limits, or specific performance rankings. It remains unknown how the model handles edge cases in network instability or specific language dialects beyond general availability status.
Primary sources
Frequently asked questions
Does the $0.05 per minute price include backend tool usage?
No. The source states that voice sessions cost $0.05 per minute, but backend model and tool usage is charged separately.
Can I use my own backend for reasoning tasks?
Yes. The documentation mentions client delegation as an option to connect your own backend for handling reasoning and tools during the live session.