OpenAI Releases GPT-Live-1.1: New Voices in 6 Languages and Fewer "checking..." Pauses

OpenAI has released GPT-Live-1.1, an update to the voice model that powers Voice mode in ChatGPT and in Codex. New voices are available for six language variants: French, German, Hindi, Mexican Spanish, Korean and Japanese. The update rolls out to all users today. The company promises API access "soon," without a date.

The announcement is short: a video on OpenAI's channel and three lines of description. It contains no benchmarks, no latency figures and no comparison with the previous version. So what follows covers only what is known, plus a few cautious conclusions drawn from it.

What changed in the voice

There are three main points: new voices, more accurate accents and fewer "checking..." pauses. The first two are self-explanatory. The third, judging by the wording, refers to moments when the voice assistant goes off to fetch information or call a tool and a filler pause hangs in the air. If those pauses are less frequent, conversation should flow more smoothly. OpenAI does not say how much less frequent.

A word on accents. For a voice interface, this is a question of trust: French speech with English phonetics, or Japanese with the wrong intonation, immediately gives away that the voice is synthetic. The explicit mention of Mexican Spanish suggests that OpenAI is working with regional variants of a language rather than a generic "Spanish."

Where it works and who gets access

The description lists Voice in ChatGPT, Codex and one more product called "dots." The text does not make clear what that is: a separate service or a typo. We won't guess.

Voice in Codex is more interesting than the list of languages itself. Codex is a coding environment, and voice there means controlling an agent by conversation: assigning a task, asking follow-up questions and getting an answer without touching the keyboard. In that scenario, pauses like "checking..." are especially irritating, because the agent already runs long operations.

Key parameters of the release:

• Model: GPT-Live-1.1.
• New voices: French, German, Hindi, Mexican Spanish, Korean, Japanese.
• Products: Voice in ChatGPT, Codex and one more item called "dots."
• Availability: all users, rollout began today.
• API: "soon," with no date or price given.
• Quality and latency metrics: not published in the announcement.

Practical takeaways

Users in the six language groups should test the update themselves: open voice mode and talk in your native language on a range of topics, from everyday questions to work terminology. Subjective accent quality and the naturalness of pauses tell you more than any line in the description. If the voice has not appeared, the rollout may not have reached your account yet.

Developers and product teams have no reason to rush. Without an API, you cannot build GPT-Live-1.1 into your own app. When it does arrive, watch three things: the price per minute of speech, the latency to first audio and how the model behaves when calling tools. That is where "checking..." lives.

For teams building localized voice products for India, Korea, Japan and Latin America, this release is a signal: a voice competitor is strengthening in exactly these markets. If you have your own voice stack, now is a good time to put together a test set of phrases in these languages so you can compare results as soon as the API appears.

And until there are official numbers, treat the wording "better" and "fewer" as a claim, not a proven fact.