Start and prioritize
“Start a Codex task to run the tests, investigate failures, and prioritize anything blocking the release.”
Codex voice mode · verified July 24, 2026
How to use GPT-Live Voice in Codex
OpenAI added GPT-Live-powered ChatGPT Voice to Codex in the ChatGPT desktop app on July 23, 2026. You can now talk to Codex to start, prioritize, interrupt, redirect, and check agent tasks while the same tools, sandbox, and approvals remain in force.
Last verified:
What changed in Codex on July 23
The new Codex voice mode is a live control and coordination layer. GPT-Live manages the natural conversation, while GPT-5.6 Terra starts and coordinates work in the desktop app. It is not a new Codex coding model and it does not replace the model, tools, or permissions used by a delegated task.
This distinction matters when people search for “GPT-Live Codex” or “OpenAI Codex voice mode.” Voice lets you direct existing Codex work conversationally; the separate GPT-Live-1 API is still represented by an official notification form, not public developer access.
Codex Voice vs dictation vs Codex CLI realtime
Choose the surface that matches the job instead of treating every microphone feature as the same product.
| Experience | What it does | Where it runs | Best for |
|---|---|---|---|
| GPT-Live Voice in Codex | A full-duplex conversation that can start, steer, interrupt, and monitor Codex tasks | ChatGPT desktop app on macOS or Windows; paired iOS Remote | Hands-free task coordination and multi-agent follow-up |
| Voice dictation | Turns speech into editable prompt text before you send it | Where the product exposes a microphone or dictation control | Writing one prompt faster without a live conversation |
| Codex CLI realtime or voice input | A terminal workflow or third-party integration with its own setup | Terminal or developer-controlled environment | Custom command-line workflows; not the July 23 desktop Voice feature |
How to use voice with Codex
1. Open a new empty Codex task
Use the current ChatGPT desktop app on macOS or Windows. The task must be empty because Voice has to be selected before the first message.
2. Select Start new voice chat
Do this before typing or sending anything. If the task starts in another mode, the microphone provides dictation rather than the GPT-Live conversation.
3. Complete first-use setup
Allow microphone access, choose a voice, and on macOS review the optional Screen context setting. A configurable Voice chat hotkey is available under Settings → Voice.
4. Speak the outcome and keep steering
Ask Codex to begin work, then interrupt naturally to reprioritize, redirect, request a progress check, or send a follow-up instruction.
Useful Codex voice commands
State the task, expected result, and next decision clearly. These prompts exercise the actions the official feature supports.
Interrupt and redirect
“Pause the deployment task. Check the accessibility regression first, then continue only if it passes.”
Coordinate agents
“Check the active research and implementation tasks, summarize blockers, and send each agent the next instruction.”
Check progress
“Tell me which Codex tasks are still running, what changed, and which decision needs my approval.”
Use project context
“Open the current project context, compare the implementation with the plan, and start a focused review task.”
Follow up
“Send the test results back to the implementation task and ask it to fix only the confirmed failures.”
Multi-agent coordination keeps Codex permissions
You can use GPT-Live in Codex to check active conversations and projects, start separate threads for longer work, coordinate multiple agents, and hear progress, blockers, or completion results. Voice can also reuse project context and supported connected tools available to the selected experience.
Voice does not bypass security controls. Every task keeps its existing tools, sandbox, network rules, and approval policy. If Codex would normally ask before internet access or a change outside the workspace, speaking the instruction does not remove that approval.
Codex voice platforms, plans, and usage limits
Desktop Work and Codex Voice has its own rolling five-hour allowance. Do not substitute the separate limits documented for ordinary ChatGPT Voice.
| Area | Current official position | What to remember |
|---|---|---|
| Platforms | ChatGPT desktop app on macOS and Windows | Standalone Work/Codex Voice is not offered on web or mobile |
| Remote access | Paired iOS Remote from a desktop host | The host must stay awake, online, and signed into the same account and workspace |
| Plans | Plus, Pro, Business, Edu, and Enterprise | Free and Go are not listed for this desktop feature |
| Voice allowance | A separate rolling five-hour window; one active Voice conversation | Plus about 15–30 minutes, Pro $100 about 1–2.5 hours, Pro $200 unlimited Voice, Business and legacy Enterprise/Edu about 45 minutes |
| Credit plans | Credit-based Business, Edu, and Enterprise use about 6 credits per minute | Delegated Codex work still consumes the normal Codex or shared usage budget |
| Enterprise and Edu | Early access initially requires Advanced voice capabilities and Early Model Access | Workspace administrators can disable Voice |
Why the Voice button may be missing
Confirm that you are in the current macOS or Windows desktop app, signed in with an eligible ChatGPT plan, and starting a new empty Codex task. Then check workspace Voice settings, app and microphone permissions, and whether another Voice conversation is already active. API-key login does not provide desktop Voice.
Open the full GPT-Live troubleshooting guideCodex voice mode FAQ
Does Codex have voice mode?
Yes. Since July 23, 2026, eligible users can start GPT-Live-powered ChatGPT Voice in Codex from a new empty task in the ChatGPT desktop app.
How do I talk to Codex?
Open a new empty Codex task on macOS or Windows and select Start new voice chat before sending the first message. Allow microphone access, then speak the task and continue with natural follow-ups.
Is Codex Voice different from dictation?
Yes. Voice is a live, interruptible conversation that can coordinate work. Dictation only converts speech into prompt text before you send it.
Does Codex voice mode work on Windows?
Yes. OpenAI lists both the macOS and Windows ChatGPT desktop apps for Voice in Codex.
Can I use Codex Voice on iPhone?
You can use it through a paired iOS Remote connected to an awake, online desktop host. OpenAI does not describe a standalone mobile Work/Codex Voice experience.
Does speaking to Codex use Codex credits?
The conversation uses a separate rolling five-hour Voice allowance. Tasks that Voice delegates still consume the normal Codex or shared usage budget.
Can Voice coordinate multiple Codex agents?
Yes. It can check active work across conversations and projects, start threads, send follow-ups, and report progress or blockers from multiple agents.
Why do I only see voice input instead of a live conversation?
The task likely started outside Voice mode. Create a new empty task and select Start new voice chat before sending any message.
Is the GPT-Live-1 API now public because Codex supports Voice?
No. The Codex feature is a ChatGPT desktop experience. The GPT-Live-1 API page still offers notification signup rather than public API access.
Official sources
OpenAI Help Center · verified ChatGPT Release NotesOpenAI · verified ChatGPT Voice in Work and CodexOpenAI · verified Pricing and desktop Voice limitsOpenAI Help Center · verified GPT-Live FAQOpenAI · verified Permission modesOpenAI · verified Remote connectionsOpenAI Help Center · verified ChatGPT Enterprise and Edu Release NotesOpenAI · verified GPT-Live-1 in the API