iiRecord
Agentic AI Atlas · OSWorld
benchmark:os-worlda5c.ai
II.
Benchmark JSON

benchmark:os-world

Structured · live

OSWorld json

Inspect the normalized record payload exactly as the atlas UI reads it.

File · benchmarks/benchmarks/benchmarks-agentic-fullstack.yamlCluster · benchmarks
Record JSON
{
  "id": "benchmark:os-world",
  "_kind": "Benchmark",
  "_file": "benchmarks/benchmarks/benchmarks-agentic-fullstack.yaml",
  "_cluster": "benchmarks",
  "attributes": {
    "displayName": "OSWorld",
    "homepageUrl": "https://os-world.github.io/",
    "kind": "full-stack",
    "targetsKind": "AgentVersion",
    "description": "OSWorld (Xie et al., 2024) is a scalable real-computer environment\nbenchmark for multimodal agents performing open-ended tasks in\nUbuntu, Windows, and macOS with full GUI/keyboard/mouse access;\n369 real-world tasks across browser, OS, and productivity apps.\n"
  },
  "outgoingEdges": [
    {
      "from": "benchmark:os-world",
      "to": "skill-area:desktop-automation",
      "kind": "covers",
      "attributes": {
        "attributes": {
          "coverage": "full",
          "weight": 0.2
        }
      }
    }
  ],
  "incomingEdges": [
    {
      "from": "eval-result:os-world.claude-sonnet-4-5.001",
      "to": "benchmark:os-world",
      "kind": "scored_against",
      "attributes": {}
    },
    {
      "from": "eval-run:os-world.claude-sonnet-4-5.2025-09",
      "to": "benchmark:os-world",
      "kind": "for_benchmark"
    }
  ]
}