OpenAI Releases Voice Conversation Model "GPT-Live-1" as Developer API

OpenAI has released a new model called "GPT-Live-1" that can process voice exchanges in real time as a developer API. The model's greatest feature is its support for "full-duplex" communication, enabling users to hear the other party's voice while speaking simultaneously. This makes it possible to implement an experience closer to natural human-to-human conversation within applications.

Most conventional AI voice systems have used a "half-duplex" method, where speaking and listening take turns alternately. This has often resulted in unnatural situations where users cannot respond until the other party finishes speaking, or cannot interrupt. GPT-Live-1 removes these constraints by incorporating a mechanism that allows responses even while the other party is still speaking, thereby positioning it as a step forward in the usability of voice AI.

In terms of performance, the model scored 80.1% on a test measuring interactivity—the naturalness and responsiveness of conversations. Compared to the previous generation model's score of 45.4%, this represents a significant improvement. It should be noted that this figure was published by OpenAI itself and is not an independent verification result from a third party.

The usage fee is set at $0.05 per minute (approximately 7 to 8 yen). Whether this price is cost-effective for business purposes varies significantly depending on the use case and call volume. For example, services like customer support or voice assistants where call durations tend to be long must consider that costs could accumulate quickly.

By being provided as an API, enterprises and individual developers can integrate GPT-Live-1 into their own applications and services. Various applications are conceivable, including voice-based customer support, language learning, and accessibility enhancement tools. On the other hand, when implementing the service, careful estimation of pricing strategy and expected user engagement time becomes crucial.

In the voice AI sector, multiple companies continue to compete, with real-time performance and improved naturalness being common objectives. The release of GPT-Live-1 attracts attention as a demonstration of OpenAI's technological position in this field. Going forward, feedback from actual users and developers is expected to become important material that will shape the direction of next-generation models.