GPT-Live

GPT-Live launched on July 8, 2026 as an architectural replacement for ChatGPT's Advanced Voice Mode, not an incremental update. The core change: where Advanced Voice Mode was turn-based, waiting for silence before processing, GPT-Live runs a full-duplex design that processes incoming audio while generating spoken output simultaneously. The model decides many times per second whether to keep talking, pause, acknowledge the user, or delegate a harder query to a more capable background model.

At launch, that delegation layer routes complex reasoning and search tasks to GPT-5.5 mid-conversation, letting the voice layer stay responsive while heavier work runs in the background. The product ships as two tiers: GPT-Live-1 for paid ChatGPT users (Go, Plus, Pro) and GPT-Live-1 mini for free users. Both rolled out globally across iOS, Android, and web on the same day, affecting more than 150 million weekly ChatGPT voice users.

For builders, GPT-Live is primarily a consumer product at launch: API access was announced as coming soon rather than available on day one. The near-term implication is that production voice agents still build on GPT-Realtime APIs rather than GPT-Live directly. The longer-term implication is that full-duplex voice is now the baseline user expectation, which raises the design bar for any product with a voice interface.