OpenAI Releases GPT-Live-1.1 with New Accents

Update adds six international accents, an English voice named Bubbie, and swaps out repetitive search audio cues for natural transitions.

OpenAI deployed GPT-Live-1.1 and a compact companion model, GPT-Live-1.1 Quip, to ChatGPT Voice on October 15, 2026. Detailed in official release notes and an announcement on X, the release addresses two day-to-day friction points: it improves voice naturalness across non-English accents and eliminates the mechanical habit of repeating “checking…” whenever the assistant runs a web query in the background.

The update has started rolling out across web browsers and ChatGPT mobile apps. The goal is straightforward: make real-time audio interaction less robotic for non-native English speakers while smoothing out awkward pauses that break conversational flow.

What Changes in GPT-Live-1.1

The updated audio rollout adds tailored voices for French, German, Hindi, Mexican Spanish, Korean, and Japanese, alongside an English voice called Bubbie. According to OpenAI documentation, distribution is rolling out in stages, meaning available options in the interface will depend on regional settings and individual account status.

The most noticeable practical change comes down to conversational timing. In earlier builds, whenever the assistant ran external searches or consulted tools, it repeated a static audio cue saying “checking…”. Version 1.1 swaps that loop for short, varied filler phrases, mimicking how a person naturally stalls for a few seconds while pulling up information.

Quip Model and API Access

Alongside the primary release, OpenAI introduced GPT-Live-1.1 Quip. This build uses a reduced parameter footprint aimed at low-latency spoken responses. Official release materials note support for Quip within standard ChatGPT apps as well as Codex and dots workflows.

Developers hoping to build on top of the new release will have to wait. OpenAI marked API access for both 1.1 models as “coming soon.” The company provided no release timetable and did not announce any changes to the baseline $0.05 per minute rate set for the original GPT-Live model.

Architecture and Usage Limits

The GPT-Live architecture arrived in July 2026, ditching the older pipeline that chained speech-to-text, a central language model, and text-to-speech engines. The current architecture processes audio end to end, listening and speaking simultaneously while routing heavy reasoning tasks to frontier models running asynchronously in the background.

Usage limits for the 1.1 models remain identical to the tier structure introduced in September. The entry-level Go plan includes 3 hours of daily access to the mini model; Plus subscribers receive 3 hours on the standard model; the $100 Pro tier provides 15 hours; and the $200 Pro plan retains unlimited real-time voice time.