AI Industry

ChatGPT Voice Gets GPT-Live-1.1 and New Hindi Voices

Oct 16, 2026·5 min read·By Rahish Saifi

Photo of a smartphone showing a glowing sound wave on a wooden table with marigold petals and a brass cup of chai

OpenAI began rolling out GPT-Live-1.1 and a smaller GPT-Live-1.1 Quip in ChatGPT Voice on Oct. 15, 2026, pairing a refreshed speech model with 13 new Live voices across French, German, Hindi, Japanese, Korean, Mexican Spanish, and English. The update lands as rivals push into real-time voice—xAI shipped Grok Voice 3.0 a day earlier—and as OpenAI expands how ChatGPT talks, searches, and remembers inside a single voice session.

Key facts

Item	Detail
Models	GPT-Live-1.1 (successor to GPT-Live-1); GPT-Live-1.1 Quip (mini successor to GPT-Live-1 mini)
Rollout	ChatGPT Voice, Oct. 15, 2026 (OpenAI on X); also updates ChatGPT, Codex, and Dots per community announcement
New voices	13 Live voices: French (Aube, Rivage), German (Sonna, Miko), Hindi (Leher, Neer), Japanese (Himari, Rindō), Korean (Haetsal, Hanul), Mexican Spanish (Tinta, Arrullo), English (Bubbie)
Tiers	GPT-Live-1.1 default for Plus and above (including Codex and Dots); Quip default for other users
Capabilities	Web search, memory, visual widgets, text and images in one chat, plugins and connected apps; no video or screen sharing in Live
API	Community post: rolling out to all users Oct. 15; API "coming soon"

What OpenAI rolled out

In ChatGPT release notes and the ChatGPT Voice help article, OpenAI describes GPT-Live-1.1 and Quip as rolling into Voice with more natural accents across languages. Live also draws on a wider range of short phrases to signal when it is searching the web, so users hear less repetitive filler while a lookup runs.

An official announcement in the OpenAI Developer Community on October 15 framed the release as "New voices. Better accents. And less 'checking…'" OpenAI said GPT-Live-1.1 is an update for ChatGPT, Codex, and dots, with considerably better performance in French, German, Hindi, Mexican Spanish, Korean, Japanese, and many other non-English languages. Delegation is more natural and steerability improved, the post said, and the rollout was targeting all users that day.

Both models are full-duplex, according to the GPT-Live-1.1 System Card dated October 15: they listen and respond continuously, following pauses, interruptions, and changes in pace. That architecture matters for voice that feels like a conversation rather than turn-based dictation.

The 13 new voices

OpenAI lists the new Live voices by language in its help documentation. The table below lists the 13 new voices OpenAI announced; the help page also lists Brazil voices Viola and Rio, which OpenAI did not count among the 13. Voices shown in your app may vary by region and account.

Language	Voice names
French	Aube, Rivage
German	Sonna, Miko
Hindi	Leher, Neer
Japanese	Himari, Rindō
Korean	Haetsal, Hanul
Mexican Spanish	Tinta, Arrullo
English	Bubbie

To choose a voice after the rollout reaches your account, open Settings → Voice → Voice in ChatGPT.

What is better

OpenAI's public messaging clusters around accent quality, less audible waiting, and more controllable dialogue. The community announcement highlights stronger performance in several non-English languages, with Hindi among the named improvements, while release notes emphasize natural accents across languages.

Live now varies the short phrases it uses while searching—less repetitive than a steady "checking…" loop—and OpenAI says delegation feels more natural with improved steerability when you redirect a session mid-thought. The system card's full-duplex design supports barge-in and pace changes that make that possible.

Live can still use web search and memory, show visual results through supported widgets, mix text and images in the same chat, and work with plugins and connected apps. It does not support video or screen sharing, per the help article.

Who gets GPT-Live-1.1 vs Quip

The system card draws a clear line by subscription tier. GPT-Live-1.1 is the default voice model for Plus and higher tiers, including in Codex and Dots. GPT-Live-1.1 Quip—described in release notes as "a mini version of the model"—is the default for other users.

OpenAI's help center lists GPT-Live-1.1 Quip for Free and Go users, and GPT-Live-1.1 for Plus, Pro, Business and other paid tiers. If you are on Plus and Voice still feels like the older Live mini, check that the rollout has reached your account; OpenAI said the update was “rolling out today to all users,” and its release notes say the voices you see may vary by region and account.

Why India and Hindi matter here

In an October 16 report, Moneycontrol quoted Vijaye Raji, OpenAI's CTO of Applications, saying voice is a key focus because India is an audio-first market, and that OpenAI is working on conversations where users switch between English and Hindi in the same session. The piece noted that GPT-Live was first introduced in July.

Dedicated Hindi voices—Leher and Neer—plus accent improvements give ChatGPT a clearer local sound and align with code-mixed Hinglish use many listeners expect.

The safety trade-off in the system card

OpenAI published a system card alongside the launch. On production-prompt voice evaluations, the company reports the new models at parity with their predecessors, with no statistically significant regressions—for example sexual content (0.77 → 0.81) and mental health (0.78 → 0.81) on the metrics OpenAI cites.

Harder synthetic-prompt evaluations tell a more mixed story. OpenAI reports statistically significant regressions for GPT-Live-1.1 versus GPT-Live-1 on sexual (0.87 → 0.84), mental health (0.70 → 0.63), emotional reliance (0.86 → 0.81), and gore (0.76 → 0.69). The card says more research is needed and that OpenAI plans to improve in future releases. It also notes these evals do not account for system-level protections such as Trusted Contact, localized crisis helplines, and Parental Controls.

Readers should treat synthetic-prompt scores as stress tests, not a forecast of everyday harm rates, but they show how more natural voice can shift refusal and reliance behavior.

How to switch voices

After the update is live for you, pick a Live voice under Settings → Voice → Voice. If a voice from the table above is missing, it may not be enabled for your region yet.

Some users in the community thread reported voice-selection glitches on Android and iOS after the rollout—issues described in user replies, not confirmed by OpenAI. If Voice fails to stick to your chosen speaker, treat those reports as anecdotal until OpenAI documents a fix.

What is next

The October 15 community post said GPT-Live-1.1 is coming soon to the API, which matters for developers building voice agents outside the ChatGPT app. Until then, the consumer Voice rollout is the reference implementation for accent quality and duplex behavior.

Inside OpenAI's product surface, Live already ties into the same ecosystem as GPT-6 Intelligent UI and Dots always-on agents—voice is one entry point into search, memory, and connected tools. Expect more language pairs and session-mixing features if India remains a stated priority.

Bottom line

OpenAI's GPT-Live-1.1 and Quip refresh ChatGPT Voice with 13 new voices, stronger multilingual accents—including Hindi—and UX tweaks that reduce repetitive search filler. Plus subscribers get the full Live model by default; others get Quip. The system card shows real-world safety parity on production prompts but flags regressions on synthetic stress tests. API access is promised next; for now, Settings → Voice is where users hear the difference.