GPT-Live-1.1 is OpenAI's November 19, 2026 update to its GPT-Live voice model. OpenAI says it "improves pronunciation, accent consistency, and transcription in multilingual voice conversations," and it adds a standby setting that pauses inference between turns. We read the changelog, the model page and the guides on November 20, 2026, and every claim below links to that source.
Key Takeaways
- OpenAI released GPT-Live-1.1 on November 19, 2026, and its changelog names pronunciation, accent consistency and multilingual transcription as the improvements.
- GPT-Live-1.1 bills $0.05 per active minute and $0.02 per standby minute, by the second, with backend model and tool usage billed separately.
- Standby applies to WebRTC sessions only, so a phone call over SIP does not get the lower rate.
- GPT-Live-1.1 and GPT-Live-1 have separate concurrent-session limits, so a limit raised for one does not carry over.
- Topcalls agents call in up to 78 languages, a ceiling that depends on the models you choose.
What did GPT-Live-1.1 change?
OpenAI changed three things on November 19. The model improves pronunciation, accent consistency and transcription in multilingual conversations. Standby, a new setting for GPT-Live-1 and GPT-Live-1.1 WebRTC sessions, pauses inference during inactivity and resumes when someone speaks. And the prompt format changed for handoffs to the backend.
The model page says GPT-Live-1.1 "can listen while it speaks" and "maintains a more consistent accent and voice across languages." It is a full-duplex model: it takes audio and text in and gives audio and text out. It runs on the Live endpoint only, not on Chat Completions, Responses or Realtime.

We found no language list on the model page, and no accuracy figure. OpenAI describes the gain in words, so any percentage you see elsewhere is not from OpenAI's docs.
Does GPT-Live-1.1 work for phone calls in other languages?
It can, through the Live API with a SIP connection. OpenAI's changelog says that since November 2, 2026, "Outbound SIP calling is now available to all organizations through the Live API," so you connect GPT-Live to a phone number through your SIP provider. Our GPT-Live API guide covers that setup.
What OpenAI does not publish is a per-language quality table. OpenAI describes the multilingual gain but gives no measurement, so test it on your own calls. Play a recorded call in the target language to a native speaker and listen for the accent, the names and the numbers.
Voices matter too. On November 2, Vapi's changelog added nine GPT-Live voices for German, French, Hindi, Japanese and Korean, so platforms already pick voices language by language.
What is GPT-Live standby, and does it apply to phone calls?
Standby is a mode where a WebRTC session stays connected but stops running inference until someone speaks. OpenAI's session guide says to set wake_policy.type to speech, and idle_timeout_s to a whole number from 5 to 60 seconds. The session keeps its ID and conversation history.
It does not apply to phone calls over SIP. The guide describes standby for WebRTC sessions, and the model page says the lower rate "applies to WebRTC sessions while inference is paused." A call on a phone line is billed at the active rate.

OpenAI also warns that after standby, "the first response to a short utterance may take longer." That trade-off suits an in-app assistant and not a live sales call.
How much does GPT-Live-1.1 cost?
OpenAI prices GPT-Live-1.1 the same as GPT-Live-1: $0.05 per active minute and $0.02 per standby minute, billed by the second, according to its pricing page. Backend model and tool usage is billed separately.
| Item | GPT-Live-1.1 |
|---|---|
| Active session | $0.05 per minute |
| Standby session (WebRTC only) | $0.02 per minute |
| Backend model and tools | Billed separately |
| Billing unit | Per second |
That is the voice model alone. A phone line, recording, transcription and campaign tooling are not in it. Topcalls AI calls start from $0.25 a minute, depending on the country you call, billed by the second, with the AI voice model, call recording, transcription and analytics included.
What should you ask a calling platform about multilingual voice?
Ask which languages work end to end, because a platform needs both a model that transcribes the language and a voice that speaks it. A big count is a ceiling. Topcalls agents call in up to 78 languages, and how many you get depends on the transcription and voice models you choose.
Ask four questions before a campaign:
- Which languages are tested on phone audio, not only in a demo.
- Which voices exist per language, and whether a regional accent is available.
- How transcripts are returned, so a native speaker can audit them.
- What the disclosure says in each language, which matters under EU AI Act Article 50.
Topcalls returns a recording, a transcript, a summary, a disposition and a post-call analysis for every call. A dedicated team configures the agent for each language, and you measure it against your own KPIs. See multi-language support, our guide to the best multilingual voice AI agents and AI voice agents.
To price a multilingual campaign, run the ROI calculator or book a strategy call.
Sources
- OpenAI API changelog: the November 19, 2026 entry for gpt-live-1.1 and the November 2 entry on outbound SIP calling.
- GPT-Live-1.1 model page and API pricing, read November 20, 2026.
- Managing GPT-Live sessions and Prompting GPT-Live.
- Vapi changelog: the November 2, 2026 entry on GPT-Live voices.
Frequently Asked Questions
Get AI calling tips in your inbox
No spam. One email per week with actionable sales automation tips.






