1. Three ways ChatGPT runs on a PC, and which one you have
OpenAI now ships a new ChatGPT desktop app that combines Chat, Work and Codex on macOS and Windows. If you used the older Windows app, it can stay installed beside the new one under the name ChatGPT Classic; OpenAI says Classic still receives model updates, bug fixes and security patches, but some new agent features arrive only in the new app. If you used the Codex app, updating it turns it into the new ChatGPT app.
On Windows both apps come from the Microsoft Store. OpenAI’s Windows page gives winget install --id 9PLM9XGG6VKS -s msstore for the new app. Classic is the app OpenAI documents for Windows 10 version 17763.0 or later on x64 or Arm64, and its companion window opens with Alt+Space, changeable under Settings > App > Companion window hotkey. Alt+Space opens a quick chat; it is not a dictation key.
Linux has no Store app, but as of October 2026 the same desktop app is available as a preview for Ubuntu, Debian, Fedora and Arch Linux. Everywhere else, chatgpt.com in a browser gives you Chat with its microphone and Voice controls.
- New ChatGPT desktop app (Windows, Linux preview): Chat, Work and Codex in one window.
- ChatGPT Classic (Windows): the previous app, still supported, with the Alt+Space companion window.
- chatgpt.com: ChatGPT in a browser on Windows or Linux.
2. Dictate into the ChatGPT composer with Ctrl+Shift+D or the microphone
In the new desktop app, OpenAI’s prompting guide says to press Ctrl+Shift+D while the composer is visible and start talking; ChatGPT transcribes your speech into the composer so you can edit it before sending. OpenAI’s feature table lists voice dictation for Plus, Pro, Business and Enterprise accounts in the app and marks it unavailable when you sign in with an API key. On chatgpt.com, the microphone icon in the message box is the dictation control: record one message, stop, and edit the text that comes back.
Both paths send audio to OpenAI. Its dictation FAQ says the recording goes to OpenAI’s models for transcription and the audio is kept as long as the chat stays in your history. After you delete the chat, the clip is deleted within 30 days, unless OpenAI must keep it for security or legal reasons or it was already shared for training and disassociated from your account. Consumer accounts that chose to share audio may have dictation audio used for training; turn off Include your audio recordings, or Improve the model for everyone, under Settings > Data Controls. OpenAI says it does not train on business users’ inputs by default.
If the microphone never starts on Windows 11, open Settings > Privacy & security > Microphone. Microsoft says Microphone access and Let apps access your microphone must be on, along with the toggle for a Microsoft Store app such as ChatGPT. A browser that doesn’t appear in that list counts as a desktop app and is covered by Let desktop apps access your microphone.
- Desktop app: Ctrl+Shift+D, speak, review, then send.
- Website: select the microphone icon, speak, stop, then edit.
- The transcript isn’t posted until you send it, but the audio has already gone to OpenAI for transcription.
3. ChatGPT Voice on Windows and the web: modes, hours and per-minute cost
Voice is the separate feature that talks back. On chatgpt.com, select the Voice icon in the prompt window, allow the browser to use the microphone, and start speaking; the microphone control mutes you and the exit control ends the call. Under Settings > Voice you may see Live (GPT-Live-1 or GPT-Live-1 mini, depending on plan), Advanced and Standard. Live can listen while it speaks, so you can interrupt it, and OpenAI warns that the transcript added to the chat afterwards is not a verbatim record.
As of October 2026, Voice in Chat is measured over a rolling 24 hours: Free gets limited GPT-Live-1 mini access, Go gets 3 hours of GPT-Live-1 mini, Plus 3 hours of GPT-Live-1, the $100 Pro plan 15 hours and the $200 Pro plan unlimited GPT-Live-1. OpenAI lists Voice in Chat on the web, iOS and Android.
In the desktop app on Windows, Voice is aimed at work: select Start voice chat or Start new voice chat to talk through a task, check progress or steer Codex and Work, and set your own key under Settings > Voice > Voice chat hotkey. OpenAI says desktop Voice is offered on Plus, Pro, Business, Edu and Enterprise plans, draws on your Codex usage budget at $0.05 a minute, and allows only one active voice chat across the app.
Audio from Live and Advanced conversations is stored with the chat transcript and retained for 30 days, OpenAI says; with Standard, the audio is deleted once it has been transcribed, unless you share audio for training.
4. ChatGPT on Linux: the desktop preview, the website and Wayland
OpenAI’s Linux page describes the desktop app as a preview for Ubuntu 24.04 LTS and 26.04 LTS, Debian 13, Fedora 43 and 44, and current Arch Linux, each on x64 or ARM64. Install the .deb with sudo apt install ./chatgpt_amd64.deb, the .rpm with sudo dnf install, or OpenAI’s script on Arch. Installation adds a signed OpenAI package repository, so later updates come through apt, dnf or pacman. Other distributions may work but are not formally supported.
OpenAI names Computer Use as a feature that is not yet in the Linux preview. The page neither lists dictation or Voice among the limitations nor confirms them, so try Ctrl+Shift+D on your own machine before relying on it. Native Wayland support is experimental: in a Wayland session the app uses XWayland when it can, and if you force native Wayland with chatgpt --ozone-platform=wayland, OpenAI warns that keyboard shortcuts, focus and floating windows may not fully work.
The website is the fallback that needs no package: open chatgpt.com in your browser, allow the microphone for the site, and use the microphone icon or Voice as you would on Windows.
- Supported: Ubuntu 24.04 and 26.04, Debian 13, Fedora 43 and 44, Arch Linux, on x64 and ARM64.
- Not yet in the preview: Computer Use.
- Wayland: XWayland by default; native mode may break shortcuts.
5. Win+H voice typing in ChatGPT: free, online and literal
Windows’ own voice typing works in the ChatGPT composer as in other text fields. Put the cursor in the box, press Windows key+H, wait for Listening…, and speak; say “Stop listening” or select the microphone to stop, and use Windows key+Alt+H to move through the menu from the keyboard. Microsoft says voice typing uses online speech recognition powered by Azure Speech, so it needs an internet connection, a working microphone and a cursor in a text box. Windows 10’s dictation has the same internet requirement.
The gear icon holds the voice typing launcher, automatic punctuation, a profanity filter and the default microphone, and Windows 11 lists 46 voice typing languages. On Copilot+ PCs, Fluid dictation corrects grammar, punctuation and filler words with on-device small language models; say “Undo that” or press Ctrl+Z if it rewrites something you meant literally. Be careful with the spoken command “Press Enter”: in ChatGPT it sends the prompt.
For dictation without a network on Windows 11, Microsoft’s Voice access (version 22H2 and later) controls the PC and writes text without an internet connection. Say “Dictation mode” first so your words are not read as commands.
6. IraVoice: on-device dictation into ChatGPT on Windows and Linux
Use ChatGPT’s own dictation when sending audio to OpenAI is fine; it is built in and already sits in the composer. IraVoice is for the other case: speech recognized on your computer, and one key that works the same in ChatGPT, your editor and your email. IraVoice 1.2.2 runs on macOS 13+ on Apple silicon, Windows 10 or 11 (64-bit), and Linux (beta). It recognizes speech with NVIDIA Parakeet; the default Parakeet Unified EN model (about 700 MB) downloads once, and after that dictation works offline.
On Windows, put the cursor in the composer of the desktop app, ChatGPT Classic or chatgpt.com, hold Right Ctrl, speak, and release. IraVoice pastes the text with Ctrl+V and then restores your clipboard. Tap the key instead of holding it for hands-free dictation, and press Esc to cancel. It does not press Enter by default, so the prompt waits for you to read it.
On Linux, IraVoice is a beta. Its AppImage runs on any 64-bit distribution from 2022 or later, Fedora and Arch included, and updates itself; a .deb and an .rpm are offered too. It is built for x64 only, so it won’t run on ARM64 machines. The default key is Ctrl+Space, and IraVoice types with wtype, kwtype, dotool, ydotool or xdotool, whichever fits your session; you may need to install one. On Wayland, where global shortcuts may not reach the app, bind a GNOME or KDE custom shortcut to iravoice --toggle-transcription, or to the AppImage’s full path followed by --toggle-transcription.
The ChatGPT app is not on IraVoice’s list of recognized developer apps, so IraVoice does not spell project code names inside ChatGPT; it does in Windows Terminal or PowerShell when you work in a Git project, for example with the Codex CLI. IraVoice does not speak ChatGPT’s replies or change how OpenAI handles a prompt once you send it. IraVoice Local costs $79 once for one computer, $109 for two, or $149 for three, with a 3-day free trial that starts at your first dictation.