GPT-Live-1.1 updates Voice in ChatGPT, Codex, and dots

OpenAI is putting a new voice model in more places than ChatGPT Voice. On October 15, 2026, the company said GPT-Live-1.1 brings new voices, better accents, and less "checking..." to Voice in ChatGPT, Codex, and dots, with new voices in French, German, Hindi, Mexican Spanish, Korean, and Japanese. The post said the update is rolling out that day to all users, and coming soon to the API. That is a later step, not a live API launch.

Two models, not one default

The October 15 release notes mark GPT-Live-1.1 and GPT-Live-1.1 Quip generally available in ChatGPT Voice. Quip is described as a mini version of the model. OpenAI says Live uses more natural accents across languages, and a wider range of short phrases to let you know when it is searching. The same-day post's "checking..." line is separate wording; the notes describe the search cue, and do not reprint that phrase.

New Live voices include options for French, German, Hindi, Japanese, Korean, and Mexican Spanish, plus Bubbie in English. The voices you see may vary by region and account. To choose a voice, open Settings > Voice, then select Voice.

The "all users" line and the tier split both stand. OpenAI's deployment safety note says GPT-Live-1.1 succeeds GPT-Live-1, and Quip succeeds GPT-Live-1 mini. GPT-Live-1.1 is the default voice model for Plus and higher-tier users, including for Codex and Dots. Quip is the default for other users. On that note, using voice with Codex or Dots does not by itself put every account on the larger model.

Both models remain full-duplex. OpenAI says they can listen and respond continuously instead of waiting for a clearly defined turn to end, follow pauses, interruptions, and changes in pace, and decide in the moment whether to respond or keep listening.

Justin Uberti posted the same day that GPT-Live-1.1 offers much better multilingual support, including several new voices, more natural delegation, and improved steerability. He said it is rolling out now to ChatGPT Voice, Desktop, and Dots. Desktop is in his post. OpenAI's post named ChatGPT, Codex, and dots, and left the API on "coming soon."

What the safety note reports

OpenAI says safeguards for these models were built largely the same way as for their predecessors: training the models to respond safely, voice-specific evaluations, existing system-level safety integrations, and the same monitoring, review, and enforcement infrastructure.

On production-prompt voice evaluations, built from opted-in user audio after privacy and eligibility checks, OpenAI says the new models perform at parity with their predecessors, with no statistically significant regressions. The note says those evaluations are not prevalence-weighted and were meant to be difficult.

On synthetic prompts aimed at hard edge cases, OpenAI says the models perform close to parity, but that GPT-Live-1.1 has statistically significant regressions versus GPT-Live-1 on sexual, mental health, emotional reliance, and gore. More research is needed, the note says, and OpenAI plans to improve those categories in future releases. Those synthetic scores do not account for system-level interventions such as Trusted Contact, localized crisis helplines, and Parental Controls, and are not a guide to safety across all production traffic.

Without delegation, OpenAI says neither model could plausibly be considered High in the Preparedness Framework's tracked categories. When the voice experience delegates to flagship models, OpenAI says the experience inherits those models' safeguards.

Open Settings > Voice and see which voices your region and account actually list. For the Plus-and-higher versus Quip default, and for the evaluation tables, read the GPT-Live-1.1 deployment safety note. Treat API access as still ahead, not available.

Two models, not one default

What the safety note reports

Sources