OpenAI Launches GPT-Live Voice Models

OpenAI released GPT-Live on July 8, a new family of voice models capable of listening and speaking simultaneously in real time, a shift from the turn-based interactions that have defined most AI voice assistants. Two versions, GPT-Live-1 and GPT-Live-1 mini, are rolling out to users globally.

The launch extends work OpenAI began in May, when it introduced three audio models for its developer platform. Those were designed to make voice-based software agents more conversational and able to complete tasks in real time. GPT-Live pushes further by allowing a model to process incoming speech while generating its own spoken response, the kind of overlapping dynamic that characterizes human conversation.

The timing arrives as OpenAI prepares for an initial public offering. The company has been expanding beyond text-based chat into audio and agentic capabilities that function in real-world settings. Voice has become a competitive front. Google, Meta, and startups like Character.AI have all invested heavily in low-latency speech models. The ability to listen and speak at the same time could position GPT-Live for customer service, live translation, and hands-free computing where delayed responses frustrate users.

OpenAI did not disclose pricing or technical benchmarks in the announcement. It remains unclear how the models perform in noisy environments or across languages. Still, the release signals that OpenAI sees voice as a central interface for the next phase of AI, one where models need to match the speed and ease of talking to another person.