Using a computer

An agent that works a computer the way a person does: opening apps, clicking, typing and moving between windows to finish a task. Good means the task is finished correctly without a person stepping in.

Sources last checked 2026-10-23.

We can't show scores for this kind of work yet. XLANG Lab, University of Hong Kong publishes results from one test. See who tests it

Key evidence rings earned, out of 4Filled: best score in that test. Dashed: tested, figures not shown here.

independent tester scores shown here, from the last year two or more testers repeat runs, some or all models

Vendor claims have not been collected for any category yet.

OSW2 ↗

ⓘ More about this category

Reliability not yet measured
1 source, linked results only
Sources last checked 2026-10-28
Ring 4: needs rings 1 to 3 first
OSW2 = OSWorld 2.0

Tested by

OSWorld 2.0
XLANG Lab, University of Hong Kong, checked 2026-10-28
Whether an agent completes long, realistic computer workflows from start to finish on a real desktop, across office files, browsers and several apps at once.
XLANG Lab reports that agents fail by losing track of constraints, missing information that arrives mid-task, guessing rather than asking, and skipping checks, rather than by stumbling on basic clicking or coding.
Last updated 2026-10-15Covers 6 of the models we trackIts authors include researchers from Alibaba Qwen and Snorkel AI.Scores not shown: permission requested on 2026-10-27
See their results
More about this test: OSWorld 2.0
Reliability not measured by this source.
Technical details for OSWorld 2.0
Binary accuracy: Whether an agent completes long, realistic computer workflows from start to finish on a real desktop, across office files, browsers and several apps at once.
License: Score reuse not stated. Checked 2026-10-28.

What this doesn't tell you

The tests run in prepared computers, not your desktop or your apps.
An agent that can operate a computer can also make mistakes on it. These tests do not measure the cost of a wrong click.
Scores from different sources are not comparable, even when they share a unit.
None of these tests were run on your own documents, code or data.
No source for this task allows us to show its figures yet. Follow the links for their results.

Data version 2026-10-28+91ac3fc9d071 · Terms of use

Using a computer

At a glance

Tested by

OSWorld 2.0

What this doesn't tell you