Benchmark family

OSWorld

A family groups the versions of one benchmark. It is not evaluated and has no applicable models of its own; each of its 7 definitions is resolved separately.

AgenticLong HorizonMultimodalComputer UseDesktop GUI

Definitions

Compare the newest 4 →

Newest first, following the recorded successor chain. Each definition is resolved against models separately.

BenchmarkChainWorld2026
ActiveNewest in family
Derived from OSWorld
Scale
347 counts
Primary metric
Task Success Rate
Applicable models
51
BenchmarkOSWorld V2.1V2.1 / 2026-09-16
Active
Successor of OSWorld 2.0 — 2026.08.08
Scale
108 tasks
Primary metric
Task Success Rate
Applicable models
51
BenchmarkOSWorld 2.0 — 2026.08.082026.08.08
Superseded
Successor of OSWorld 2.0 — 2026.06.24
Scale
31 tasks
Primary metric
Task Success Rate
Applicable models
51
BenchmarkOSWorld 2.0 — 2026.06.242026.06.24
Superseded
Successor of OSWorld-Verified
Scale
27.25 checkpoints
Primary metric
Task Success Rate
Applicable models
51
BenchmarkOSWorld-VerifiedVerified
Superseded
Successor of OSWorld
Scale
361 counts
Primary metric
Task Success Rate
Applicable models
51
BenchmarkOSWorld2024
Superseded
Scale
43 tasks
Primary metric
Task Success Rate
Applicable models
51
BenchmarkOSWorld-Pro2026
Active
Related to OSWorld
Scale
67,264 counts
Primary metric
Task Success Rate
Applicable models
51

Related outside this family

TUA-Bench — derived from — OSWorld

Family methodology

No family-level methodology recorded.

Across definitions

Tasks: Computer Use, Desktop Gui Interaction, Long Horizon Computer Use, Multi Application Workflow
Sources: None recorded at family level
Last reviewed: 2026-09-23

Applicability does not mean evaluated; benchmark result/model score data is outside this UI contract.