OpenAI Brings GPT-Live's Full-Duplex Voice Control to Codex and the ChatGPT Desktop App: Ushering in the Era of Hands-Free Agentic Coding

OpenAI has introduced GPT-Live's full-duplex voice technology in combination with Codex and ChatGPT Work into its desktop applications for macOS and Windows. This technology is a model that can listen while speaking at the same time, eliminating the rigid turn-taking of conventional conversations and enabling natural interaction.

With this update, software engineers can direct multi-threaded coding tasks, review Pull Requests, and debug applications using voice commands alone. This creates a 'hands-free' development environment where developers can coordinate their work without typing or switching windows.

Key features include the following:

• Separation of the real-time voice layer from the execution engine: While GPT-Live maintains natural conversation, a reasoning model such as GPT-5.5 running in the background handles complex computational tasks.
• Context awareness: Through the Appshots feature in the macOS desktop app, ChatGPT analyzes the currently active window, local files, codebase structure, and plugins to perform pair-programming.
• Multitasking execution: A single voice prompt can simultaneously trigger multiple tasks such as bug investigation, API migration review, and unit test generation.

Additionally, developers can convert design mockups into code, and through remote execution via iOS, check on agent progress and redirect tasks without switching applications.