OpenAI GPT-Bidi-1 leak points to simultaneous voice input-output

OpenAI is reportedly testing an unannounced voice model called GPT-Bidi-1 that can speak and listen at the same time, according to Android Authority on code references and early user tests.

TL;DR

• Android Authority reported that OpenAI is testing an unannounced bidirectional voice model named GPT-Bidi-1.
• The reported bidirectional design would let ChatGPT speak, hear, and process interruptions simultaneously during a conversation.
• The report said the model had started rolling out to a small group of app users ahead of a possible wider release.

Android Authority reported that GPT-Bidi-1 appeared in code references first identified by TestingCatalog. The report said the model name suggests a bidirectional voice system and described it as part of OpenAI's next voice upgrade for ChatGPT.

According to the report, GPT-Bidi-1 can speak, hear, and listen simultaneously, which would allow mid-sentence interruptions to be handled more naturally. Android Authority said TestingCatalog found internal code describing the model as a "major leap in intelligence" and "the next generation of Voice."

Android Authority also reported that GPT-Bidi-1 had already started rolling out to a select group of ChatGPT app users. The report did not cite an official OpenAI announcement, but said the limited rollout suggested a possible release window later that week.

Related questions

• What is GPT-Bidi-1?
• What did Android Authority report about OpenAI's new voice model?
• How does bidirectional voice interaction differ from current voice assistants?
• Has OpenAI officially announced GPT-Bidi-1?