OSWorld
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
OSWorld (885-sample split)
98.08Eligible Rate
15
OSWorld-G refined annotation
83.5Text Match
14
OSWorld (Verified)
78.7Success Rate
13
OSWorld (test)
38.54Success Rate
12
OSWorld (Verified)
75Score
12
OSWorld 361 tasks
54.65Overall Success Rate
10
OSWorld 50 easy tasks 1.0 (test)
100ASR
10
OSWorld
28.2Accuracy
9
OSWorld
60.1Accuracy
8
OSWorld-Verified (Class-Imbalanced, Human Evaluation) 1.0 (test)
88.5Precision
7
OSWorld Verified Class-Imbalanced Test Scripts 1.0 (test)
61.9Precision
7
OSWorld Verified Class-Balanced Human Evaluation 1.0 (test)
94Precision
7
OSWorld Verified Class-Balanced Scripts 1.0 (test)
79.2Precision
7
OSWorld-G-R
72.34Accuracy
6
OSWorld
0.73Pearson r
6
OSWORLD (test)
36.7ASR
6
OSWorld (test)
42.5Success Rate
6
OSWorld subset
46.48Success Rate
5
OSWorld Chrome
93.5Reward Accuracy
5
OSWorld
92Reward Accuracy
5
OSWorld 361 tasks
52.1Success Rate
5
OSWorld
20.3AUV
5
OSWorld w/o Loop
29.5AUV
5
OSWorld w/ Loop
6.9AUV
5
OSWorld overall
86.8Overall Success Rate
4