Browser OS
Build a tiny desktop operating system in the browser: wallpaper, taskbar and clock, and five apps that open in real windows you can drag, maximize, minimize and close.
Results
12 setups · the middle run of eachsort
Press “+ Compare” on two results to put them side by side.GPT-6.1 Solmedium19/19
cost $0.24time 7m17s
Claude Sonnet 5.5high19/19
cost $0.95time 9m53s
Claude Opus 5.5medium19/19
cost $2.30time 15m54s
Claude Haiku 5.5low18/19
cost $0.026time 3m15s
GPT-6.1 Solhigh18/19
cost $0.23time 12m10s
Claude Opus 5.5high18/19
cost $3.82time 19m19s
Claude Haiku 5.5max0/19
cost $0.26time 34m07s
Anthropic models run in Claude Code 2.1.296 · OpenAI models run in Codex CLI 0.162.0 · no harness of our own (yet) · why it matters
What the robot checks
The same 19 items for every model, written from the prompt. Scores are always out of 19, even when a page breaks halfway.
- Opens without a page error
- Taskbar and clock visible
- Animated wallpaper
- Right-click menu on the desktop
- Double-click opens Notes
- Double-click opens Calculator
- Double-click opens Files
- Double-click opens Game
- Double-click opens Settings
- Window drags by its title bar
- Maximize fills the screen
- Minimize hides the window
- Close removes the window
- Clicking a window brings it to the front
- Calculator: 12 + 30 = 42
- Notes survive a reload
- No uncaught exception during the session
- No console errors
- No network attempts
The prompt
Sent as is to every model. Test v1 · checks v4: any change makes a new version.
Build a browser-based desktop operating system as ONE self-contained HTML file.
Hard rules
Read the whole prompt
Build a browser-based desktop operating system as ONE self-contained HTML file. Hard rules - Everything lives in a single HTML file: inline CSS and JavaScript only. No external URLs of any kind: no CDNs, web fonts, remote images or network requests. The file will be opened offline, straight from disk. - It must run in a current desktop Chrome at 1920×1080 with zero console errors. The desktop 1. A full-screen desktop with an animated, procedurally drawn wallpaper (CSS or canvas, no image files). 2. A taskbar with a start/menu button, one button per open window, and a live clock. 3. Desktop icons that open five apps: Notes, Calculator, Files, Game and Settings. 4. Real windows: drag by the title bar, resize from an edge or corner, minimize, maximize/restore and close. Clicking a window brings it to the front. 5. A right-click context menu on the desktop with at least two working actions. The apps - Notes: a text editor whose content survives a page reload (use localStorage). - Calculator: + − × ÷, decimals and clear. Pick either operator precedence or left-to-right evaluation, and be consistent. - Files: a fake folder tree with at least 3 folders and 5 files; opening a text file shows its content. - Game: a small playable game of your choice, with a score. - Settings: switch between at least two themes or wallpapers. Testing contract (an automated tester relies on these attributes, so they must be exact) - Taskbar element: data-taskbar. Clock element: data-clock. - Desktop icons: data-launch="notes", data-launch="calculator", data-launch="files", data-launch="game", data-launch="settings". A double-click on an icon opens its app. - Each app window's root element: data-window="<the same name>". Its title bar: data-titlebar. Its buttons: data-action="minimize", data-action="maximize" and data-action="close". - Calculator buttons: data-key="0" … data-key="9", data-key=".", data-key="+", data-key="-", data-key="*", data-key="/", data-key="=" and data-key="C". Calculator display: data-display. - Notes text area: data-notes-input. Make it look polished and coherent, like a real product. Output format: reply with the complete file inside a single ```html code block and nothing else.
Credit. Inspired by Bijan Bowen's "Browser OS Test". Prompt and checks by Nerdict.