Voice models stop waiting for your turn: GPT-Live-1.1 goes full-duplex as the default for Plus and above
Synopsis
The GPT-Live voice family has moved to GPT-Live-1.1 and GPT-Live-1.1 Quip, both keeping the full-duplex design that lets a model listen and respond continuously, follow pauses, interruptions and pace changes, and decide in the moment whether to answer or keep listening. GPT-Live-1.1 succeeds GPT-Live-1 as the default voice model for Plus and higher tier users, including Codex and Dots, while GPT-Live-1.1 Quip succeeds GPT-Live-1 mini as the default for others. Safety work follows largely the same path as the predecessors: training models to respond safely, measuring performance on voice-specific evaluations, reusing existing system-level safety integrations, and using the same monitoring, review and enforcement infrastructure.
Interpretation
Voice conversation models no longer work around a clearly defined turn boundary; GPT-Live-1.1 and GPT-Live-1.1 Quip can listen and respond continuously, follow pauses, interruptions and changes in pace, and decide in the moment whether to respond or keep listening. The earlier GPT-Live-1 and GPT-Live-1 mini were also full-duplex, so the new versions are iterations within the same family rather than the first introduction of full-duplex behavior. The system card places both models as the most recent iterations in the GPT-Live family and states that their full-duplex capability matches the predecessors.
GPT-Live-1.1 succeeds GPT-Live-1 as the default voice model for Plus and higher tier users, including Codex and Dots, while GPT-Live-1.1 Quip succeeds GPT-Live-1 mini as the default voice model for others. The default voice model role moves from the earlier GPT-Live-1 and GPT-Live-1 mini to the new versions. The system card states each new model's predecessor and the user range for which it is the default.
Safety work follows largely the same path as the predecessors: training models to respond safely, measuring performance on voice-specific evaluations, reusing existing system-level safety integrations, and using the same monitoring, review and enforcement infrastructure. Compared with the predecessors, the safety build does not start a separate track but stays largely the same. The system card states that these practices are largely the same as those for the predecessors of the GPT-Live-1.1 models.
Perspective
The system card covers the positioning and safety description of the latest GPT-Live family iterations, spanning GPT-Live-1.1 and GPT-Live-1.1 Quip and the user ranges for which they are the default voice models. It suits researchers and engineers following voice interaction products and safety engineering.
The system card says the safety approach is largely the same as the predecessors, but it does not give specific performance figures on voice-specific evaluations, nor does it say how much the new versions change naturalness or intelligence relative to the predecessors.