OpenAI Computer-Using Agent. SOTA benchmark results across computer control tasks. OSWorld 38.1% (human: 72.4%), WebArena 58.1%, WebVoyager 87%.
Metrics
Info
ProviderOpenAI
Organization that created the model
Licensecommercial
Software license type
Agent Typecomputer_use
Agent specialization area
Base Model Requirementsgpt-4-vision, o1
Required foundation models to operate
Other
Osworld Score38.1
Webarena Score58.1
Webvoyager Score87
Score Breakdown
adoption
15%90
Community usage, market traction, and ecosystem maturity
tool use
94
memory context
88
self reflection
90
planning reasoning
92