# GPT-Live-1.1 adds 13 voices, and the system card notes safety regressions

> OpenAI is rolling out GPT-Live-1.1 and Quip to ChatGPT Voice, Codex, and dots, with 13 new voices. The system card reports regressions on sexual, mental health, emotional reliance, and gore evals. The API is not out yet.

Published 2026-10-15T08:00:00-07:00 · Model & Plan Comparisons · Canonical: https://news.littlewolflabs.com/posts/2026-10-15-gpt-live-1-1-voices

On October 15, at 11:19 AM PT, [OpenAI](https://x.com/OpenAI/status/2110797830361436604) posted GPT-Live-1.1: new voices, better accents, and less of the "checking…" filler. It is rolling out the same day to Voice in ChatGPT, Codex, and dots, for all users. The API is "coming soon."



The [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) name the new Live voices as French, German, Hindi, Japanese, Korean, and Mexican Spanish, plus Bubbie in English. [OpenAI's forum announcement](https://community.openai.com/t/gpt-live-1-1-is-rolling-out-with-new-voices-and-better-accents/1406704) calls that set 13 new voices. The [Voice help page](https://help.openai.com/en/articles/20001274-chatgpt-voice) lists two in each of those languages (Aube and Rivage, Sonna and Miko, Leher and Neer, Himari and Rindō, Haetsal and Hanul, Tinta and Arrullo) and Bubbie, and also lists Viola and Rio for Brazil. The release notes do not call the Brazil pair new. The voices you see can vary by region and account.

The [GPT-Live-1.1 system card](https://deploymentsafety.openai.com/gpt-live-1-1/introduction) says GPT-Live-1.1 is the default for Plus and higher, including Codex and Dots. GPT-Live-1.1 Quip, the successor to GPT-Live-1 mini, is the default for everyone else. On synthetic voice evals, the card says GPT-Live-1.1 has statistically significant regressions versus GPT-Live-1 on sexual, mental health, emotional reliance, and gore, and that OpenAI plans to improve those categories. The same section says the production-prompt evals showed no statistically significant regressions, and that the synthetic set is not a guide to safety across real traffic. The card also notes slight regressions on a "respecting warnings" eval. The models are meant to be used under OpenAI's [Usage Policies](https://openai.com/policies/usage-policies).



[OpenAI's forum post](https://community.openai.com/t/gpt-live-1-1-is-rolling-out-with-new-voices-and-better-accents/1406704) repeats the rollout line, "today to all users" and "coming soon to the API." Later posts on that thread report voice-selection bugs since the rollout: on Android, Standard and Advanced ignoring the voice chosen in settings, and on iOS, Standard switching to Live. Those are user reports, not an OpenAI confirmation.



## What it costs

The announcement does not mention a price change. The [Voice help page](https://help.openai.com/en/articles/20001274-chatgpt-voice) still lists the existing allowances, including limited GPT-Live-1.1 Quip on Free, 3 hours of Quip on Go, and 3 hours of GPT-Live-1.1 on Plus. Audio clips from Live stay with the transcript for 30 days, under the [Privacy Policy](https://openai.com/policies/privacy-policy).

If you need a voice API this week, the GPT-Live API is not the one that is live. [Grok Voice 3.0](/posts/2026-10-14-grok-voice-3) is on the Voice Agent API at $0.08 a minute.

## Should you switch

The new voices are on the account you already pay for. Read the regression note before you use Live for sexual, mental health, emotional-reliance, or gore-adjacent conversations. If Standard or Advanced is speaking in the wrong voice after the update, the forum thread is the report to watch.