OpenAI introduced GPT‑Live, the latest voice model powering ChatGPT’s conversational experience across mobile apps and web browsers. Unlike the previous turn‑based system, GPT‑Live operates on a full‑duplex architecture that allows the assistant to hear and respond at the same time. Users now hear brief, natural interjections—"mhmm," "yeah"—while they speak, reducing awkward pauses and cut‑offs that plagued earlier versions.
The new model runs on a single multimodal engine, handling speech‑to‑text conversion, language generation and text‑to‑speech synthesis in one pipeline. When a question demands deeper reasoning or research, GPT‑Live can silently hand the task to more capable back‑end models such as GPT‑5.5, keeping the conversation flowing without noticeable lag.
Access to GPT‑Live varies by subscription tier. ChatGPT Pro, Plus and Go users receive the full‑featured GPT‑Live‑1 model, while those on the free plan are limited to GPT‑Live‑1 mini, a lighter version without the customizable intelligence settings. To start a voice session, users tap the waveform icon at the right edge of the text box, grant microphone permission if prompted, and a floating orb appears on the screen. The orb remains active even if the user switches apps or locks the device.
Paid subscribers can adjust the model’s “intelligence” level—Instant, Medium or High—through the settings menu. Higher levels add a split‑second of extra processing time, allowing the model to perform more thorough reasoning before replying. The mini model does not support this customization. Users can also select different voices, change language settings, or set voice mode as the default launch option.
While GPT‑Live marks a significant step forward, it does not yet support screen or video sharing. Users wishing to revert to the older voice model can do so by navigating to the Voice settings and selecting a previous option. The upgrade, however, has already earned praise for its fluidity. Reviewers note that the assistant no longer interrupts mid‑sentence, and the conversational rhythm feels almost anticipatory, especially during longer queries.
Beyond casual chats, GPT‑Live shows promise for live translation tasks. A transcript of the spoken exchange is always available by scrolling down, and the model can generate interactive widgets when a visual answer is appropriate. The default Instant intelligence level handles everyday questions quickly, while more demanding topics benefit from the Medium or High settings, which invoke deeper analysis without breaking the natural flow.
Overall, GPT‑Live positions ChatGPT as a more approachable voice assistant for users who prefer speaking to typing. By merging listening and speaking into a seamless loop and offloading heavy reasoning to background models, OpenAI aims to make AI conversations feel less like interacting with a script and more like talking to a knowledgeable partner.
This article was written with the assistance of AI.
News Factory APP - agentic news to boost your SEO & AEO.