OpenAI Ships a Linux ChatGPT Desktop Preview—but the Browser Still Wins for Now

OpenAI Ships a Linux ChatGPT Desktop Client—and the First Cut Falls Short

OpenAI has released a preview of ChatGPT Desktop for Linux, extending its native desktop apps to every major operating system. The client is a proprietary Electron program distributed as DEB and RPM packages, with stated support for Ubuntu 24.04 LTS and 26.04 LTS, Debian 13, and Fedora 43 and 44 on both x86-64 and Arm64 machines. It combines three increasingly overlapping products: ChatGPT for general assistance, ChatGPT Work for delegated workplace tasks, and Codex for software development.

ZDNET's first look on Linux Mint found the package underwhelming. When asked for technical specifications for the SpaceX and Nvidia Starmind AI datacenter satellites—using a link to the relevant site—the app returned a height of 20 meters and wing width of 70 meters, while SpaceX's stated figures are 30 meters and 75 meters. For a reviewer who primarily uses AI as an advanced search engine, that kind of factual miss is difficult to overlook.

Performance and integration problems compounded the disappointment. The idle app consumed more than a gigabyte of RAM, and after a few hours of light use on a 16GB system it leaked enough memory to push the machine into virtual memory and near-unusable slowdowns. ChatGPT shell jobs did not appear in the desktop project list, and early Wayland reports include broken Japanese and Korean Fcitx5 input on Fedora 44 KDE Plasma unless the app is launched with specific Chromium flags.

The Linux release arrives roughly a month after Anthropic introduced a Linux desktop version of Claude. OpenAI says Linux was one of its most-requested desktop platforms, but the current DEB/RPM distribution leaves Arch Linux, NixOS and users seeking Flatpak or AppImage packages waiting for now.

What OpenAI's Linux Push Really Means for Codex and the Developer Desktop

A Developer-Desktop Land Grab, Not Just a Detached Browser Window

For OpenAI the strategic prize is not another ChatGPT interface; it is Codex running inside a local project workspace. On Linux, developers could already use ChatGPT in a browser, but many relied on shell use, API integrations, unofficial Electron wrappers and editor extensions. A native client that can examine and modify files, run commands inside selected directories, display repository changes and handle diff editing gives OpenAI a more direct route into the environments where developers actually build AI infrastructure and applications.

The Claude Comparison Arrives Early

Anthropic's Linux Claude desktop client shipped before OpenAI's preview and supports a somewhat broader baseline—Ubuntu 22.04 or later and Debian 12 or later, against OpenAI's Ubuntu 24.04/26.04, Debian 13 and Fedora 43/44. A cited ZDNET developer survey also found that 75% of surveyed developers prefer Claude Code over Codex. Entering that market late with a rough preview, narrower distro support and no Flatpak or AppImage option gives Claude an early opening among Linux users.

Electron Explains the Resource Drain

The desktop app's over-1GB idle footprint and memory leak are consistent with its Electron architecture. Electron combines Chromium and Node.js to ship one codebase across Windows, macOS and Linux, but it is generally slower and more memory-hungry than native applications. For Linux power users and developers who prize efficiency, that engineering trade-off—alongside the client's closed-source license—may be a stronger barrier than any missing feature.

Privacy and Permissions Define the Trust Boundary

The client does not automatically scan disks, control arbitrary Linux GUI applications or gain root access. Connected services such as Google Calendar, Gmail, Outlook, Teams, Slack, Google Drive, Figma, GitHub and Adobe require separate approval, and native computer use is not available in this preview. At the same time, consumer accounts are set by default to use conversations to improve OpenAI's models unless the user disables "Improve the model for everyone," and anything deliberately shared is sent to ChatGPT for processing. For security-conscious Linux users, these defaults make careful setup as important as the app's coding features.

Sensible Setup and Safeguards for Linux Users Testing the ChatGPT/Codex Preview

• Keep "Improve the model for everyone" switched off under Settings > Data Controls unless you deliberately want your consumer chats used for model improvement; your history remains but is not used for training when the toggle is off.
• Use temporary chats for sensitive one-off queries, accounting for the stated 30-day deletion window, and avoid granting access to ~/, your entire documents tree, or directories containing credentials and source files.
• Give Codex a dedicated, narrow project directory and leave its permission mode on "Ask for approval"; do not switch to "Full access" unless a specific task genuinely requires it.
• Run the app with a separate browser profile that only holds the service accounts needed for the task, and avoid keeping password managers, private email, banking sessions or confidential source material open at the same time.
• If you are on Arch Linux, NixOS or another unsupported distribution, or you need reliable Wayland input, wait for broader packaging or launch with flags such as --ozone-platform=wayland and --enable-wayland-ime; official guidance says native Wayland support is experimental and the app uses XWayland when available.
• Before relying on the desktop client for terminal workflows, confirm that your shell jobs appear in the project list—ZDNET's review found they did not—and keep an eye on memory use, since the preview consumed over 1GB at idle and leaked memory after a few hours.