Introducing GPT‑Live
OpenAI has introduced GPT‑Live, an upgraded voice mode model for ChatGPT that can delegate complex tasks to GPT‑5.5 in the background while maintaining conversation flow. The author had preview access and notes the new model is a significant improvement over the previous GPT‑4o-based voice mode, though they encountered an early bug where the model inappropriately laughed at non-joke statements.
Background
- OpenAI's "GPT‑Live" is a major upgrade to the voice mode in the ChatGPT mobile app, replacing the older GPT‑4o‑based model that had been stagnant since 2024.
- The new model can delegate complex tasks (web search, deep reasoning) to OpenAI's latest "frontier model" (currently GPT‑5.5) in the background, while continuing to chat with the user — so the conversation doesn't pause while the heavy lifting happens.
- Simon Willison is a well-known British programmer and AI blogger; his personal site (simonwillison.net) is widely followed for detailed, hands-on commentary about LLM releases.
- The post highlights a quirky bug: the model kept interrupting with laughter at ordinary statements, which highlights the ongoing challenge of tuning AI models' social behavior (making them sound natural without being offensively condescending).
- Voice mode is a key competitive front between OpenAI and rivals like Google (Gemini Live), Anthropic, and others; improving latency, naturalness, and capability is critical for user adoption beyond text chat.