OpenGPTEvidence-based model decisionsAgent configuration · OSWorld 2.0
OSWorld Agent · Claude Opus 4.8 · Max · Batched tool
The ranked object: model, scaffold, tools, permissions, budget, and environment together.
System identity
Agent systemOSWorld Agent · Claude Opus 4.8 · Max · Batched tool
Base modelClaude Opus 4.8
Source model ID or labelClaude Opus 4.8
ScaffoldOSWorld 2.0 computer-use agent
Scaffold versionOSWorld 2.0
ToolsBatched computer-use tool
Comparability grouposworld-2-v2026-06-24-500
Evaluation configuration
Source snapshotOSWorld 2.0 · Task version v2026.06.24 · 500-step budget · Jun 25, 2026
Tasks108 long-horizon computer workflows
EnvironmentDesktop and operating-system workflows in reproducible virtual machines
Budget500 steps
Evaluation date
Data captured
Binary completion20.6%
Partial score54.8%
Published evidence
Open sourceEvidenceListed by the benchmark publisher
SourceOSWorld 2.0
CostNot reported
Cost basisThe source does not report a comparable cost for this configuration
Comparable cost is not reported for this configuration, so it is not treated as a low-cost option.
What this result does not prove
This result applies to the exact system and source conditions shown here. It does not establish a universal best agent, production reliability, data governance, or performance on a different benchmark version.