GPT-Live-1 (API)
2026-09-10 · Publisher : OpenAI · Domain : Artificial Intelligence
OpenAI opens GPT-Live-1 in the API: a full-duplex voice model that listens and speaks at the same time, priced at $0.05 per minute, with reasoning delegated to a text model.
What changes
- A single model handles listening and speaking, with no transcription, reasoning and synthesis chain.
- Delegation of reasoning and tool calls to a backend model (GPT-6 Astra, Luna, or a third-party model).
- Style, pace and tone controllable through the system prompt; better handling of noise and silences.
- Telephony support for voice agents per call.
- Native transcriptions and response text, keyword biasing.
### Specifications - Audio and text input and output; images not supported. - Knowledge cutoff: July 31, 2025. - Availability: API (v1/live/sessions). - Identifier: gpt-live-1. Use of the backend model is billed separately. ### Pricing | Item | Price | | --- | --- | | gpt-live-1 voice session | $0.05 per minute, billed per second | | Backend model and tools | Billed separately at the pricing of the chosen model | ### Benchmarks | Measure | Result | | --- | --- | | Full Duplex Bench, gain vs GPT-Realtime-2.1 | +30 points | | Interruptions at Speak, vs turn-based systems | about -80% | Scores published by the vendor. The post also reports a #1 ranking on Tau3 with Astra as the backend. ### Key takeaways The cost of the voice layer becomes predictable (per minute) and independent of the reasoning model, which can be chosen according to the difficulty of the case.
Links
Tags : Audio, OpenAI, Voix, Full-duplex, API