OpenAI brings GPT-Live full-duplex voice to Codex and ChatGPT desktop

OpenAI brings continuous voice to the development environment

OpenAI has announced the integration of the GPT-Live audio model, with full-duplex capability to listen and speak simultaneously, directly into the desktop ChatGPT application for macOS and Windows. The move comes two weeks after the model's initial launch on July 8, 2026, and now also powers the agentic systems Codex and ChatGPT Work — separate experiences within the same app. The change eliminates the need for push-to-talk interaction, allowing developers to converse with the AI continuously while working.

What changes in practice for developers

Technically, GPT-Live operates as a continuous audio model, capable of processing speech and generating responses simultaneously without artificial pauses. In the context of Codex, this means a programmer can dictate code snippets, ask project-related questions, and receive corrections in real time, all using voice as the primary input. The reduced latency and conversational fluency make the tool much closer to a virtual programming partner, especially useful for repetitive tasks or accessibility scenarios.

Impact on scale and enterprise adoption

For companies relying on fast development cycles, full-duplex voice integration reduces friction between intent and action. Instead of switching between keyboard, mouse, and chat interfaces, the developer maintains spoken context, accelerating iterations and reducing interruptions. From a growth perspective, the feature can increase user retention on the ChatGPT desktop platform and boost Codex adoption in engineering teams, as the cognitive cost of use drops. Funnel data indicates that voice interfaces tend to increase usage frequency by up to 40% in productivity tools.

10Dobro take: automation and performance in layers

At 10Dobro Prod, we see this evolution as another step toward systems where human interaction becomes invisible and continuous. For those operating automation and growth marketing at scale, GPT-Live in Codex represents an additional orchestration layer: it is possible, for example, to trigger deploys, adjust campaign parameters, or query real-time metrics just by using voice commands integrated with AI agents. Audiovisual work remains one sector where this can be applied — script editing, brief generation, transcript adjustment — but the main focus is on reducing operational latency and increasing parallel execution capacity. Full-duplex voice interfaces are, above all, a competitive advantage for teams that need to scale without multiplying headcount.