OpenAI announced on Thursday (November 19) that GPT-Live-1.1 is available in the API, under the model ID gpt-live-1.1 and the endpoint v1/live/sessions. It is the same voice model that has powered ChatGPT Plus and Pro voice since October 15 — now in the hands of developers building their own real-time voice agents.
What changes over GPT-Live-1
According to the official documentation, 1.1 transcribes speech more accurately, keeps accent and voice more consistent across languages, captures exact letters and numbers better, and completes tasks more reliably. Delegation also got simpler: instead of special markers, the developer describes in plain language when the model should hand reasoning or tool use to a backend model. OpenAI has not published benchmark numbers for these gains, so they should be read as vendor claims until independently measured.
Standby: silence gets 60% cheaper
The big practical addition is standby, a new billing state for WebRTC sessions on GPT-Live-1 and GPT-Live-1.1. The developer sets an idle timeout of 5 to 60 seconds without a transcript (idle_timeout_s); once it elapses, inference pauses and the session costs $0.02 per minute, versus $0.05 per minute while active — 60% less. The WebRTC connection, session ID and conversation history are kept, and the session returns to active as soon as someone speaks (wake_policy of type speech). Billing is per second, with no rounding up to the full minute, and backend model and tool usage is billed separately.
How the architecture works
GPT-Live-1.1 is a full-duplex model: it listens while it speaks, so the caller can interrupt mid-sentence. The model handles the conversation, while heavy reasoning and tool calls go to a backend model chosen by the developer, via delegation. The default context window is 128,000 tokens, with streaming and function calling supported; structured outputs, fine-tuning and predicted outputs are not supported. GPT-Live-1 and 1.1 have separate concurrent-session limits, and switching models requires a new session. For anyone already running GPT-Live-1, migration is essentially swapping the model name in the session configuration.
Image: Neumann U87 condenser microphone in a studio (photo: Will Fisher, CC BY-SA 2.0, via Wikimedia Commons; illustrative image).
Sources
- OpenAI — GPT-Live-1.1 Model (documentação oficial da API)
- MarkTechPost — OpenAI Ships GPT-Live-1.1 for Voice Agents, Cuts Idle Cost to $0.02/Min
- Apidog — What Is GPT-Live-1.1?
Earlier Geekiki coverage: OpenAI libera GPT-Live-1 na API: voz full-duplex a US$ 0,05 por minuto · OpenAI lança GPT-Live-1.1 com 13 novas vozes e sotaques melhorados no ChatGPT
Transparency note: this article was produced with the assistance of artificial intelligence, cross-checked against public posts on X and internet sources, and reviewed before publication. Please consult the original sources to confirm the information.
By Muse