
GPT-Live-1.1 API: Full-duplex voice model for multilingual agents
In brief
OpenAI released GPT-Live-1.1 via API, a full-duplex voice model with better accent consistency, transcription accuracy, and task delegation for building voice agents. Includes standby mode to pause inference during silence at lower cost.
How it works
Developers use model ID gpt-live-1.1 in Live sessions. The model listens while speaking, handles interruptions, and delegates complex tasks to a backend model via plain-language prompts. Standby mode pauses inference after configurable idle timeout (5-60s) while keeping WebRTC connection open.
Developers building voice agents that need natural multilingual conversation, precise detail handling, and reliable task completion.
Availability
Sources
5 merged into this card- 1GPT-Live-1.1 has arrived in the APIAnnouncement · community.openai.com · 20.11
- 2Managing GPT-Live sessionsHow-to guide · developers.openai.com
- 3Prompting GPT-LiveHow-to guide · developers.openai.com
- 4GPT-Live-1.1 ModelHow-to guide · developers.openai.com
- 5OpenAI API changelog — November 19, 2026Release notes · platform.openai.com · 19.11
More from OpenAI
GPT-6.2 Sol and Astra added to Codex model catalogs
OpenAI added GPT-6.2 Sol and Astra to the model catalogs in Codex 0.171.0, with Sol set as the default selection.
Remote Sessions in codex exec
OpenAI's Codex CLI now allows users to start or resume noninteractive sessions on an existing app server via the new `codex exec --remote <ADDR>` flag.