SKU RLE-1106 · Sold by External
OSWorld
Product specifications
| SKU | RLE-1106 |
|---|---|
| Data type | RL environments |
| Volume | 369 tasks (execution-based; 134 unique evaluators) |
| Size on disk | Not published by source |
| Format | Python + downloadable VM images (Docker / VMware / VirtualBox); JSON task configs |
| Access model | PUBLIC LICENSE |
| Pricing | Free · open-source license |
| Quality score | — |
| License | Apache-2.0 |
OSWorld is a scalable real-computer environment for multimodal agents that operate a full operating system (primarily Ubuntu, also Windows and macOS) across real web and desktop applications (Firefox, LibreOffice, GIMP, GNOME Files, VS Code) and OS file I/O, including multi-application workflows. Agents issue raw keyboard/mouse control via pyautogui (click, type, scroll, hotkeys) and observe screenshots, the accessibility (a11y) tree and terminal output, with optional set-of-marks aiding. It provides 369 tasks, each with an initial-state config and an execution-based evaluation script (134 unique evaluation functions).