Latest in AI
GPT-Live-1.1: New Voices, Quip, Usage Limits and API Availability
OpenAI’s voice update improves accents and conversation flow, but the model you get and the features available still depend on where you use it.
OpenAI began rolling out GPT-Live-1.1 on October 15, with new voices, more natural accents and less repetitive search updates across ChatGPT, Codex and dots. There is also a smaller model, GPT-Live-1.1 Quip. For people who speak to AI regularly, the useful change is how the conversation feels between answers.
What changed in GPT-Live-1.1
The release notes highlight new voice options for French, German, Hindi, Japanese, Korean and Mexican Spanish, alongside Bubbie in English. Voice availability can vary by account and region. OpenAI also says Live now uses a wider variety of short phrases while searching, rather than repeatedly giving the same status response.
That sounds like a small detail until you use voice for a longer task. Repeated acknowledgements take up space in the conversation, especially when you are trying to explain something unfinished. More natural accents could also make the experience easier to stay with in a language other than English. These are practical reasons to try the update, rather than evidence that every answer has become more accurate.
GPT-Live-1.1 versus Quip
According to the system card, GPT-Live-1.1 replaces GPT-Live-1, while Quip succeeds GPT-Live-1 mini. The full model is the default for Plus and higher-tier users, including Codex and dots; Quip is the default for other users.
Both remain full-duplex. They can listen while speaking and decide whether to continue responding or leave room for the person to talk. This capability existed in the previous generation, so it should not be mistaken for a new feature introduced with 1.1.
The distinction matters when comparing experiences. Two people can receive the same voice update while using different underlying models. A voice’s name alone does not tell you which model or allowance is behind it.
ChatGPT Voice limits and how to use it
Open Settings, then Voice, to select the available voice experience and voice. The current ChatGPT Voice guide lists these allowances for Voice in Chat over a rolling 24-hour period:
| Plan | Model | Voice in Chat allowance |
|---|---|---|
| Free | Quip | Limited, variable |
| Go | Quip | 3 hours |
| Plus | GPT-Live-1.1 | 3 hours |
| Pro, $100/month | GPT-Live-1.1 | 15 hours |
| Pro, $200/month | GPT-Live-1.1 | Unlimited, subject to abuse guardrails |
These are Chat allowances. Desktop voice has separate pricing and limits, and Work tasks also count toward normal Work usage. Workspace settings and regional availability can affect access.
Live still does not support video or screen sharing. The guide directs users to the older Advanced experience for supported mobile video and screen-sharing features. Better conversation quality therefore does not make Live a complete replacement for every existing voice workflow.
Is GPT-Live-1.1 available through the API?
OpenAI’s launch announcement says API access is coming soon. The consumer rollout should not be read as confirmation that developers can already select the new model in their own applications.
For an existing voice product, that means waiting for the actual API release, its model identifiers, pricing and migration guidance before planning a switch. A demonstration in ChatGPT does not establish the cost or behaviour of a separately deployed application.
Early replies in the developer-community announcement thread report voice-selection problems in Standard and Advanced modes on mobile. These are individual reports, not a confirmed failure rate or proof that the new Live model caused them. They do make it worth checking that your chosen mode and voice are actually active after the update.
What to watch beyond the new voices
OpenAI’s system card reports parity with the predecessors on its production-prompt safety evaluations, but also identifies regressions in several categories on targeted synthetic tests. Those tests are not estimates of everyday failure rates. They are a reminder that a more comfortable voice and more reliable judgement are separate things.
For teams evaluating voice interfaces, the useful test is whether people can explain, interrupt and correct a task with less effort while still understanding what the assistant actually did. A pleasant voice helps people keep talking. Clear evidence and controllable actions determine whether that conversation becomes useful work.
I’m Paras. I work at the intersection of AI, engineering and distribution.
I help technical founders turn product launches and complex engineering updates into clear stories that developers and founders want to read and share.