Editorial illustration for OpenAI adds voice control to desktop Codex and ChatGPT
OpenAI Adds Voice Control to Desktop ChatGPT
OpenAI is putting a voice on Codex. Two weeks after launching GPT-Live, its full-duplex audio model that can listen and talk at the same time, the company has folded that same technology into the ChatGPT desktop app on macOS and Windows. That means Codex and ChatGPT Work, both housed inside the desktop client, now run on a conversational layer that doesn't force the rigid back-and-forth of older voice assistants. Heavy reasoning still gets handed off to background models like GPT-5.5, but the front end can respond mid-sentence, interrupt, and pick up context the way a person would on a call.
The practical upshot: engineers can talk their way through multi-threaded coding jobs, ask for a pull request review, or troubleshoot a bug out loud instead of typing it out. An OpenAI spokesperson told VentureBeat this marks the first time voice activation has shipped natively with Codex on desktop, a detail that matters given Codex now counts more than 5 million weekly active users and has grown well past its original coding-assistant scope into something closer to a general productivity tool. OpenAI illustrated the feature with a demo video of two of its own engineers talking to the same session at once.
Software engineers can initiate multiple concurrent task threads from a single spoken prompt. For instance, a developer preparing to ship a feature can instruct the system to investigate an open authentication bug, review a pending API migration pull request, and generate missing unit tests simultaneously.
Why this matters
For developers, the interesting part isn't the voice gimmick, it's the architecture decision behind it. Splitting the conversational layer from the execution engine means Codex can keep chewing on a build or a refactor while GPT-Live handles the back-and-forth, dropping in "got it" style acknowledgments instead of freezing up or talking over you. That's a real engineering problem OpenAI is solving, not just a demo trick. If it works reliably, it changes how people actually work with agentic coding tools, less alt-tabbing to type follow-up instructions, more talking through a problem while the agent grinds in the background.
We'd hold off on calling this hands-free coding solved, though. Full-duplex voice layered onto agentic execution is exactly the kind of thing that looks great in a controlled walkthrough and gets messy the moment latency spikes or the agent needs clarification mid-task. Worth watching whether Codex users actually adopt voice for real sessions or whether this stays a novelty layered on top of ChatGPT Work. The decoupling idea, though, is probably where other agent builders end up too.
Related Reading
Browse all OpenAI news →
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
Related Topic OpenAI hires 630 ex-Meta staff as ChatGPT memory may turn data into ads
Business & Startups
Related Topic Sen. Warren urges Trump admin to expand AI credit after OpenAI letter
Business & Startups
From Archives Florida Pastor Sues OpenAI Over Alleged ChatGPT Medical Advice
From Archives Free ChatGPT Users Get Worse Health Advice From Older AI Model
From Archives Apple Sues Ex-Executive After 24 Years, Alleges He Took AI Secrets to OpenAI
Policy & Regulation
From Archives OpenAI's Georgia Data Center Project Secures 3.2-Gigawatt Power Deal
Business & Startups
Common Questions Answered
How does GPT-Live's full-duplex audio technology improve upon older voice assistants in ChatGPT desktop?
GPT-Live enables simultaneous listening and talking without the rigid back-and-forth interaction of traditional voice assistants. This conversational layer allows the system to provide acknowledgments like "got it" while continuing to process tasks in the background, preventing freezing or talking over the user.
What concurrent tasks can software engineers initiate through Codex voice control with a single spoken prompt?
Software engineers can instruct the system to perform multiple simultaneous actions such as investigating open authentication bugs, reviewing pending API migration pull requests, and generating missing unit tests all from one voice command. This multithreading capability allows developers to accomplish complex workflows without switching between manual inputs.
How does OpenAI's architecture decision to split the conversational layer from the execution engine benefit developers?
By separating the conversational layer from the execution engine, Codex can continue processing builds or refactors while GPT-Live handles the back-and-forth dialogue. This architectural approach solves a real engineering problem by allowing continuous work on code tasks without interruption from voice interactions, rather than serving as merely a demonstration feature.
What background models does OpenAI use for heavy reasoning tasks in the desktop ChatGPT application?
OpenAI hands off heavy reasoning tasks to background models like GPT-5.5, which work behind the scenes while the conversational layer manages user interactions. This division of labor ensures that complex computational work doesn't interfere with the real-time voice interaction experience.
Further Reading
- OpenAI debuts ChatGPT Voice so you can have ongoing conversations, ask the AI to complete tasks while hands-free - Fortune
- Building with ChatGPT Voice - OpenAI
- Introducing GPT-Live - OpenAI
- OpenAI Adds Mobile Remote Control for Codex, with Voice Interaction in Development - KuCoin
- OpenAI Adds Mobile Remote Control to Codex, Voice Interaction in Development - Phemex
