GPT-Live-1
Released Jul 8, 2026
OpenAI's full-duplex voice model: it listens and speaks at the same time and hands deeper reasoning and tool use to a separate backend model. It powers ChatGPT Voice and has been in the API since September 2026.
- Price per minute
- $0.050/min
- Inputs
- audio, text
- Tool calling
- Yes
- Languages
- —
- Voices
- 21
- Big Bench Audio
- 90.1%
- Access
- Generally available
- Licence
- Proprietary
Strengths & weaknesses
Strengths
- Full duplex: handles pauses, interruptions and backchannels while it speaks
- Delegates reasoning and tools to a backend model you choose, e.g. GPT-6 Astra
- Flat $0.05 per minute, billed per second
- 21 built-in voices plus custom voices from a prompt or recording
Weaknesses
- Backend model and tool calls are billed separately on top of the per-minute rate
- No image or video input on the API
- Served only on the new Live endpoint (v1/live/sessions), not the Realtime API
Notable: frontier-lab clause (also #3 on AA Speech to Speech Index at 83.2). releaseDate is the GPT-Live system card's publication date (2026-07-08), when GPT-Live-1 and GPT-Live-1 mini became the ChatGPT Voice defaults; API general availability was 2026-09-10 per the OpenAI changelog. openai.com pages returned 403 to the fetcher, so launch-post facts come from the OpenAI developer community announcement, the changelog and the system card. AA's Big Bench Audio score (90.1%) is for 'GPT-Live-1 (Astra, medium)', i.e. paired with GPT-6 Astra at medium effort; with GPT-6 Sol (low) AA lists 89.0%. OpenAI's launch claims (paired with GPT-6 Astra medium): Tau3 83.6% vs 45.7% for GPT-Realtime-2.1, AA Conversational Dynamics 97.3%, turn-taking latency 0.798s. Voices: 12 at API launch, 9 more added 2026-10-16 (21 in the voice table). GPT-Live-1 mini (ChatGPT free-tier default) has no API listing and no separate entry. Languages: no total published; the voice table covers English, Portuguese, German, French, Hindi, Japanese and Korean.
Links
API model strings
- openai
gpt-live-1