GPT-Live-1.1 and Quip in ChatGPT Voice: Six New Language Voices and Usage Limits for Every Plan

OpenAI is rolling out GPT-Live-1.1 and a mini version called GPT-Live-1.1 Quip in ChatGPT Voice, with more natural accents across languages and new voices for French, German, Hindi, Japanese, Korean and Mexican Spanish. The company posted the change in its ChatGPT release notes on October 15, 2026. The same day, its ChatGPT Voice help page lists a usage limit for every plan, which is the part most paying users will care about.

Plan and limit details always matter to readers, so this post collects them in one place, adds the history of the model, and marks what OpenAI has not said.

Quick reference

What OpenAI announced

The release note is short. OpenAI says it is "rolling out GPT-Live-1.1 and GPT-Live-1.1 Quip, a mini version of the model, in ChatGPT Voice, with more natural accents across languages." It adds that Live "also uses a wider range of short phrases to let you know when it's searching," and lists the new voices. To change a voice, open Settings, then Voice, then Voice. The voices you see "may vary by region and account."

That is the full announcement. OpenAI did not publish benchmark numbers for 1.1 in the notes we read, did not describe a retraining, and did not say whether the API model changes. Anything beyond the above, including claims that the update is faster or has lower latency, comes from users and should be read as impressions. One thing we can say is how Live is described: the help page says Live "can listen and speak at the same time, making turn-taking and interruptions feel more natural," and can use web search and memory, show visual results in widgets, accept text and images, and use plugins.

The usage limits, plan by plan

Two speech bubbles passing a green bead, picturing the turn-taking of full-duplex ChatGPT Voice with GPT-Live-1.1

The help page gives these limits for Voice in Chat, measured over a rolling 24-hour period:

Voice in Work follows your plan's Voice limits and billing, and Work tasks also count toward normal Work usage. Voice in the desktop app has separate pricing and limits.

Two details are worth noting. First, the Enterprise usage-based price of $0.05 per minute matches the API price OpenAI gave for the voice layer when GPT-Live-1 reached the API in September, which suggests the same pricing logic across surfaces, though the page does not say so. Second, Free and Go users get the mini model only, which is the clearest plan-based split in this release.

We cannot tell from OpenAI's pages whether these limits are new, raised or unchanged compared with before the 1.1 release, since the earlier version of the help page is not preserved there. If you track plan changes, compare against your own records or the Wayback Machine.

A short history of GPT-Live

GPT-Live is OpenAI's full-duplex voice line. The model appeared in ChatGPT in July, which we covered in GPT-Live for ChatGPT Voice, together with GPT-Live-Transcribe and GPT-Transcribe. On September 10, OpenAI launched GPT-Live-1 in the API at $0.05 per minute for the voice layer, saying it improved Full Duplex Bench performance by 30 percentage points over GPT-Realtime-2.1 and could delegate reasoning to a backend such as GPT-6 Astra. On September 23, Live gained plugin support, which we covered in ChatGPT Voice plugins and Work. The 1.1 release is the next increment.

The competition is moving the same way. Anthropic expanded Claude voice mode with tools and 11 languages, and xAI launched Grok Voice 3.0. Language coverage and accent quality are now a feature race, not a side note.

Language voices: what the help page lists

Beyond the default nine named voices, the help page lists additional Live voices by language: two each for Brazilian Portuguese (Viola, Rio), French (Aube, Rivage), German (Sonna, Miko), Hindi (Leher, Neer), Japanese (Himari, Rindo), Korean (Haetsal, Hanul) and Mexican Spanish (Tinta, Arrullo), plus Bubbie in English. The release note names the six languages other than Brazilian Portuguese as new. Treat the exact pairing of voice and language as subject to the help page changing.

The page also says you can set a preferred language under Settings, then Voice, then Language, and that choosing your most-used language can help ChatGPT understand your speech more accurately. You can also ask it mid-conversation to speak another language. It lists no accuracy figures for any language.

Which Voice mode should you pick?

The help page describes three options under Settings, then Voice, and the right choice depends on the job.

A practical test: run the same five questions in Live and Standard, including one with a pause mid-sentence and one with background noise, and note which one interrupts you. If you work in one of the six new languages, repeat the test with the matching voice and compare it with the older voices on the same prompts. Because OpenAI has published no per-language quality numbers, your own comparison is the only data you will get.

Limits and caveats worth knowing

• Live is not for groups. OpenAI says it is designed primarily for one-on-one conversation and may respond when people speak to each other.
• Interruptions still happen. The page admits that background noise, long pauses or audio from another speaker can cause them, and suggests headphones or iPhone Voice Isolation.
• No video or screen sharing in Live. Use Advanced on mobile for those.
• Transcripts are not verbatim. OpenAI says they may not exactly match what was said.
• Approvals are on screen. When a plugin action needs approval, spoken approval is not supported; you review it on screen.
• Accuracy. OpenAI reminds users that ChatGPT can make mistakes, especially on date-sensitive or location-sensitive questions.
• Teens. Voice access for a linked teen account can be managed through parental controls.

Privacy: what happens to your audio

Per the help page, audio clips from Live and Advanced conversations are stored with the transcript in your chat history and retained for 30 days. Deleting a chat deletes its associated clips within 30 days, unless OpenAI needs to keep them for security, safety or legal reasons. OpenAI says it does not train on audio or video clips unless you choose to share them, through Data Controls, though transcripts may be used if Improve the model for everyone is on, depending on plan and settings. If you use Standard mode, audio is deleted after transcription. For comparison, see our guide to Claude's voice data opt-in settings.

What this means for you

• Free and Go users get the mini model, which is a quality trade-off OpenAI does not quantify. If voice quality matters, expect a gap and test it.
• Plus users have 3 hours per rolling day, which is plenty for casual use but not for all-day voice work.
• Heavy users should look at the $100 and $200 tiers or credit-based business plans, where the 1.25 credits per minute rate sets a cost per minute.
• Non-English speakers gain native-sounding voices in six languages, but OpenAI gives no accuracy or accent benchmarks, so judge by trying them.
• Developers should not assume the API changed. The release note covers ChatGPT; check OpenAI's API changelog before updating production voice agents.

What to watch

• An OpenAI post or system card with measured differences between 1.0 and 1.1.
• Whether the API gets 1.1 and the new voices, which the notes do not promise.
• Whether the limits above change as usage grows.
• Independent tests of the Quip mini model against the full model.

Related reading

• GPT-Live for ChatGPT Voice (July)
• GPT-Live-1 in the API
• ChatGPT Voice plugins and Work
• GPT-Live-Transcribe and GPT-Transcribe
• Claude voice mode: tools and 11 languages
• Grok Voice 3.0 speech-to-speech

Details are accurate as of October 15, 2026, based on OpenAI's release notes and help pages, which can change without notice.