In brief
- OpenAI updated its GPT-Live voice model to 1.1, with 13 new voices.
- Default for Plus and higher tiers, and GPT-Live-1.1 Quip for everyone else. The API is coming soon.
- Hebrew is not on the list of improved languages, and nothing was published about its quality.
OpenAI announced GPT-Live-1.1, an update to its voice model that adds 13 new voices and better performance in non-English languages.
AI Desk · Published October 16, 2026
First look: this report is based on OpenAI’s official announcement and system card and has not yet been tested hands-on by Maitz.
OpenAI announced the update on October 15, 2026, and published the system card the same day on its Deployment Safety Hub. The company says it is rolling out to all users.
What did OpenAI announce with GPT-Live-1.1?
The update covers ChatGPT, Codex and dots. According to OpenAI, it brings 13 new voices and considerably better performance in French, German, Hindi, Mexican Spanish, Korean, Japanese and other languages.
Delegation, which hands a task to a backend model or agent mid-conversation, feels more natural. Steerability also improved. OpenAI adds that voice improvements will now ship continuously.
| Model | According to the system card |
|---|---|
| GPT-Live-1.1 | Default for Plus and higher tiers, including Codex and dots. Successor to GPT-Live-1 |
| GPT-Live-1.1 Quip | Default for other users. Successor to GPT-Live-1 mini |
| Type | Full-duplex voice models: they listen and respond continuously, with no rigid turn-taking |
| API | “Coming soon.” GPT-Live-1 is available in the API today |
What does the system card say about safety?
The system card says the new models perform close to parity with their predecessors. Many measures show no statistically significant regression.
There are exceptions. GPT-Live-1.1 shows statistically significant regressions on sexual content, mental health, emotional reliance and gore compared with GPT-Live-1. OpenAI says more research is needed.
Without delegation, neither model is considered High in any tracked category of its Preparedness Framework.
Why does a voice model update matter?
Voice mode is becoming a main way to work with models. Better accents and non-English performance decide whether users choose to talk instead of type.
For developers, GPT-Live-1 is available in the API today at $0.05 per minute, billed per second. No price has been published for 1.1 yet.
What we don’t know yet
- When the API opens and at what price. GPT-Live-1 costs $0.05 per minute, billed per second, with backend model and tool usage billed separately.
- Per-language benchmark results beyond the list OpenAI named.
What it means for you
- ChatGPT users: open voice mode and try a new voice in Settings › Voice.
- Developers: you can build on GPT-Live-1 in the API today and move to 1.1 once it is announced. For OpenAI’s related Decisions API, see our story.
- Teams: check how voice data is handled under your plan before rolling voice out to staff. Our ChatGPT guide covers settings.
- For comparison: a rival voice model is covered in Grok Voice 3.0.
More on Maitz
Sources
- OpenAI: GPT-Live-1.1 is rolling out with new voices and better accents, community.openai.com (Announcements), October 15, 2026
- OpenAI Deployment Safety Hub: GPT-Live-1.1 system card, deploymentsafety.openai.com, October 15, 2026
- OpenAI API Changelog: GPT-Live 1, developers.openai.com
- Verified: October 16, 2026
