Key takeaways

  • OpenAI released GPT-Live-1.1 on November 19 with better pronunciation, accent consistency and transcription across languages.
  • Live sessions cost $0.05 a minute when active and $0.02 a minute on the new WebRTC standby, as of November 19, 2026.

OpenAI released GPT-Live-1.1 on November 19, an update to the voice model behind its Live API, according to the API changelog. OpenAI says GPT-Live-1.1 pronounces words better, keeps a more consistent accent and transcribes speech more accurately in conversations that switch between languages. The model page adds that it can listen while it speaks, so a user can cut in without waiting for it to finish.

The same day, OpenAI added standby for WebRTC sessions, the browser-based connection most voice apps use. A session on standby pauses the model while nobody is talking and wakes up when someone speaks. That changes the bill. A Live session costs $0.05 a minute while active and $0.02 a minute on standby, as of November 19, 2026, with backend model and tool calls billed separately. A support line or a voice assistant that sits open for long stretches now pays less than half the rate for the quiet parts.

Switching is a one-line change: developers set gpt-live-1.1 in an existing Live session configuration. It takes audio and text in and out, supports streaming and function calling, and runs only on the Live endpoint, not Chat Completions or Responses. GPT-Live-1 and GPT-Live-1.1 have separate limits on concurrent sessions, so moving traffic doesn't eat into the old model's allowance.

The update follows a run of Live API additions. On November 2, outbound phone calls over SIP opened to every organization, and on October 16 OpenAI added nine voices in languages including German, French, Hindi, Japanese and Korean. GPT-Live-1.1 targets the same multilingual use: apps that talk to customers in more than one language. OpenAI has published no benchmark numbers for the new model, so the accuracy gains are its own description for now.

Sources

  1. API changelogOpenAI
  2. GPT-Live-1.1 model pageOpenAI
  3. Live conversations: configure standbyOpenAI