OpenAI Launches Full-Duplex Voice Model GPT-Live

On July 9, OpenAI officially launched GPT-Live, a new full-duplex voice model now available in ChatGPT, marking a major upgrade in AI voice interaction. The model supports natural, simultaneous listening and speaking with the ability to be interrupted anytime.

Key Details and Versioning

A core breakthrough of GPT-Live is the separation of voice conversation from heavy computational tasks. When encountering complex problems requiring search or deep reasoning, the model offloads the task to a frontier model (GPT-5.5 at launch) in the background while keeping the voice conversation uninterrupted. Additionally, it can generate real-time UI interfaces, displaying information via visual cards for weather, stocks, and sports. Regarding versions, GPT-Live-1 serves as the default for Go, Plus, and Pro users, while GPT-Live-1 mini is available for free users. OpenAI also previewed API access for both versions and opened notification registrations for developers and enterprises.

Performance and Product Evolution

According to data shared by @testingcatalog, GPT-Live-1 significantly outperforms the previous Advanced Voice Mode across multiple dimensions, including GPQA, BrowseComp, and internal τ³-Voice Telecom tests. The model was initially known as "GPT Bidi 1" before being officially renamed GPT Live 1 to unify the voice product line's naming convention. The company also showcased new capabilities like image interaction during recent live demonstrations.

2026-07-07 ~ 2026-07-09 · 44 related posts

5 near-duplicate retellings: OpenAI · BLCNYY · 赛博禅心 · AGI Hunt · bowenc0221