OpenAI Launches GPT-Live-1.1 Update with Voice and New Multilingual Voices
OpenAI announced that GPT-Live-1.1 brings updates to the voice features of ChatGPT, Codex, and dots, including new voices, improved accents, and reduced "checking..." prompts.
The new voices cover French, German, Hindi, Mexican Spanish, Korean, and Japanese; it is being rolled out to all users today and will soon be available on the API.
GPT-Live-1.1 is a successor to GPT-Live-1, utilizing a full-duplex architecture that allows simultaneous listening and speaking, handling pauses, interruptions, and rhythm changes; Plus and above users will use this model by default, while other users will use the Quip version.
Previously, GPT-Live was launched in July, supporting deep reasoning delegated to backend models while maintaining uninterrupted conversations; the voice layer in the API is billed at $0.05 per minute.
System cards indicate that the new model is on par with its predecessor in native voice evaluation, with no significant regression.
In market mechanics, the buyers are ChatGPT and Codex users who utilize voice interaction, while the seller is OpenAI's full-duplex voice model; the event is driven by model iteration, transitioning from old voice modes to new voices and better accent experiences, benefiting multilingual users and scenarios requiring natural conversation, while putting pressure on marginal features that still rely on old modes.
This update expands full-duplex voice from the basic version to multilingual voices and reduces confirmation prompts.
Source: Public Information
ABAB AI Insight
OpenAI launched the GPT-Live full-duplex voice model in July 2026, replacing the Advanced Voice Mode, and opened the API in September; this GPT-Live-1.1 adds multilingual voices and optimizes accents and confirmation behavior in ChatGPT, Codex, and dots.
The capital path reflects the separation of the voice layer as an independent front end from the backend reasoning model, expanding user engagement duration and scenarios through more natural multilingual voices while maintaining a separate billing structure for the voice layer.
This is similar to the historical transition of voice assistants from turn-taking to full-duplex continuous dialogue, and resembles the approach of model providers enhancing product coverage through language and accent expansion; it is currently in a stage of iteration from basic full-duplex to multilingual and low confirmation prompts.
Essentially, this represents a technological replacement: full-duplex voice models begin to replace old voice modes that rely on explicit turns, shifting conversations from waiting to interruptible and parallel processing, thereby reducing interaction delays and expanding the range of supported languages.
ABAB News · Cognitive Laws
Full-duplex turns waiting into interruptible
Multilingual voices retain users better than a single accent
Reducing confirmation prompts equals increasing conversation density