Model WarsJuly 8, 2026via MarkTechPost
OpenAI Releases GPT-Live and GPT-Live-1 mini: Full-Duplex Voice Models That Delegate Deeper Reasoning to GPT-5.5
Why it matters
OpenAI is shipping a new voice architecture that fundamentally changes how AI assistants interact in real-time. Full-duplex capability (simultaneous listen/speak) removes latency friction and mirrors natural human conversation—a meaningful capability leap that could reshape voice-first AI adoption.
Key signals
- GPT-Live: full-duplex voice model architecture (listen + speak simultaneously)
- GPT-Live-1 mini: lightweight variant released
- Delegates reasoning workloads to GPT-5.5
- Now powering ChatGPT Voice
- Release date: July 8, 2026
The hook
Full-duplex voice just changed the game. OpenAI's GPT-Live listens and speaks simultaneously—while GPT-5.5 handles the thinking.
OpenAI released GPT-Live, a new generation of voice models now powering ChatGPT Voice. It uses a full-duplex architecture to listen and speak at once, and delegates search and reasoning to GPT-5.5.