OpenAI Launches GPT-Live-1.1 and Quip with New Voices Across Six Languages

OpenAI has officially rolled out GPT-Live-1.1 and GPT-Live-1.1 Quip, its newest generation of real-time voice models, across ChatGPT Voice, Codex, and dots. Aimed at refining direct vocal interactions, the update introduces 13 new voices spanning six non-English languages, curtails hesitation delays during web searches, and delivers noticeably more authentic pronunciation and natural accents.

New voices. Better accents. And less “checking…”

GPT-Live-1.1 brings these updates to Voice in ChatGPT, Codex, and dots, with new voices in French, German, Hindi, Mexican Spanish, Korean, and Japanese.

Rolling out today to all users. Coming soon to the API. pic.twitter.com/L6K8ZUC1xm

– OpenAI (@OpenAI) October 15, 2026

Under this rollout, the flagship GPT-Live-1.1 model becomes the default conversational engine for paid subscribers on Plus and higher tiers, while free-tier users receive the lightweight GPT-Live-1.1 Quip. According to technical documentation, gradual deployment is underway across desktop and mobile platforms, with developer access via an API slated for release in the near future.

New Voices Across Six Languages in ChatGPT

The update expands ChatGPT’s voice options with support for more languages. Release notes published in mid-October list two new German voices, “Sonna” and “Miko”, as well as new voices for French, Hindi, Japanese, Korean, and Mexican Spanish. English speakers also gain an additional voice option called “Bubbie.”

These additions replace the legacy lineup of nine standard English-centric personas, including Breeze, Cove, and Arbor. Subscribers can switch between voices by opening Settings and navigating to the Voice section within the app, though OpenAI notes that specific voice availability may vary based on account type and geographical region.

The update also reduces repeated uses of “checking…” during web searches, replacing them with a wider range of short phrases.

How Does the New Voice Mode Work?

At its core, GPT-Live-1.1 relies on full-duplex communication architecture, allowing the system to listen and speak concurrently without awkward cutoffs. According to the technical documentation, the model can handle pauses, user interruptions, and changes in speaking speed.

According to OpenAI’s system card, GPT-Live operates as a dual-model system: the voice model handles real-time audio input and conversational streaming, while complex logic, reasoning, and computational tasks are routed to dedicated thinking models. The system card confirms that the voice models possess no direct code-execution capabilities, maintaining strict operational safeguards.

Voice Usage Limits Across ChatGPT Subscription Tiers

OpenAI has structured daily voice allowances according to subscription tier, calculated across a rolling 24-hour window. The allocations break down as follows:

• Free Accounts: Capped access to the lightweight Quip model, with no fixed hourly limit published in official documentation.
• Go Plan: Grants three hours of voice conversation with the Quip model per 24-hour cycle.
• Plus Subscription ($20/month): Includes three hours of direct dialogue powered by the full GPT-Live-1.1 model every 24 hours.
• Pro Plan ($100/month): Raises the usage ceiling to 15 hours with the full model.
• $200 Advanced Tier: Eliminates hourly time constraints entirely, retaining only automated safeguards to prevent abuse.

This tiered rollout builds on updates introduced in late September, when OpenAI enabled voice mode to interact directly with calendar and reminder apps, requiring on-screen confirmation taps before messages are dispatched.

Safety Evaluations Reveal Regressions Across Four Metrics

OpenAI’s safety report found regressions in several areas, despite improvements in audio quality and response speed. While production benchmarks matched the performance of previous models, synthetic adversarial testing showed statistically significant safety regressions in four sensitive categories.

These regressions surfaced in evaluations covering mental health prompts, emotional reliance, violent content, and sexual material. OpenAI stated that it plans to remediate these vulnerabilities in subsequent updates, advising developers and teams building sensitive voice applications to maintain independent safety guardrails.

Early Rollout Glitches and Anticipation for the API

Some Android and iOS users have reported technical issues since the rollout. Users have reported problems saving their selected voices, as well as occasional speech-recognition issues with some accents.

On the developer side, anticipation remains high for the upcoming API release. However, OpenAI has yet to publish concrete figures on per-token pricing, real-world latency, or Word Error Rates (WER). Until OpenAI publishes pricing and performance data, developers may want to wait before moving existing applications to the new model.

What Does the New Update Actually Deliver?

Ultimately, GPT-Live-1.1 brings a noticeable refinement to spoken human-computer interaction, especially for speakers of the six newly supported languages, thanks to smoother cadence and reduced verbal filler.

For end users, the decision to upgrade depends largely on practical requirements and budget. Casual users will find the free Quip model sufficient for daily conversations without added cost, while power users requiring extended, uninterrupted sessions will benefit from the higher quotas in Plus and Pro. Moving forward, OpenAI’s key challenge will be resolving early mobile bugs and addressing the safety regressions identified in its evaluation benchmarks.