OpenAI just shipped what might be the most significant upgrade to ChatGPT’s voice capabilities since the feature first launched. GPT-Live-1, which went live on July 8, is now the default voice model for ChatGPT Plus and Pro subscribers, with a smaller variant called GPT-Live-1 mini available to free-tier users.

The headline feature is full-duplex audio processing. In English: the model can listen to you and talk back at the same time, much like an actual human conversation. Previous versions operated more like a walkie-talkie, where one side had to finish before the other could respond.

What full-duplex actually changes

The model continuously processes audio input even while generating its own speech output. This means it can pick up on cues that you want to interject, adjust its pacing, or backtrack when you signal confusion. It’s a fundamentally different architecture from the Advanced Voice Mode it replaces.

GPT-Live-1 can delegate complex reasoning and task management to GPT-5.5, OpenAI’s more powerful model, while keeping the conversation flowing naturally. The model also now supports visual aids during voice interactions, including weather updates and sports scores, and maintains compatibility with existing features like memory and file uploads.

How it stacks up against Advanced Voice Mode

GPT-Live-1 was strongly preferred over Advanced Voice Mode in head-to-head conversations lasting five to ten minutes, evaluated across multiple metrics by human assessors. The model showed clear advantages in conversational flow, turn-taking, and interruption management.

On more technical benchmarks, GPT-Live-1 also outperformed its predecessor on GPQA, which measures scientific reasoning capability, and BrowseComp, which evaluates agentic web search performance. It additionally showed gains in telecom support evaluations.

Reports emerged in June about a project called GPT-Bidi-1, which appears to have been the internal codename or precursor to what became GPT-Live-1. That followed advancements in OpenAI’s GPT-Realtime model back in May, suggesting this launch is the culmination of several months of focused voice AI development.

OpenAI has also built dedicated safety measures into GPT-Live-1, including real-time safeguards for risks like self-harm conversations and emotional reliance. Global rollout across iOS, Android, and web platforms has begun, with API access also in the pipeline.