GPT-Live: OpenAI's Human-Like Voice AI Challenges Rivals

OpenAI has launched GPT-Live, a new voice model designed to make conversations with ChatGPT feel more natural. This update began rolling out on July 9, 2026. The model uses a "full-duplex" architecture, allowing it to listen and speak simultaneously. This feature helps it keep pace with competitors like Google's Gemini Live. GPT-Live can now process audio continuously, enabling natural interruptions and back-and-forth banter. It even uses conversational cues like "mhmm." OpenAI says the model makes hundreds of interaction decisions per second. This is a significant change from the older "turn-based" Advanced Voice Mode. What's interesting is that GPT-Live can also handle complex tasks. It hands off difficult questions to GPT-5.5 in the background while continuing to chat naturally. Users can also adjust the "Intelligence" level for reasoning. Internal benchmarks show a dramatic improvement. In blind tests, users preferred GPT-Live-1 over the old system 75.7% of the time. It also scored much higher in accuracy for expert-level science reasoning and complex web searches. The bottom line is that AI voice interactions are becoming much more sophisticated and human-like.