{
  "aliases": [],
  "authors": [],
  "details": {
    "benchmark_protocol": {
      "actions": [
        "agent-defined Windows interactions"
      ],
      "counts": [
        {
          "basis": "computed",
          "evidence_url": "https://raw.githubusercontent.com/microsoft/WindowsAgentArena/HEAD/src/win-arena-container/client/evaluation_examples_windows/test_all.json",
          "population": "evaluation_examples_windows/test_all.json",
          "qualifier": "exact",
          "unit": "task definitions",
          "value": 154
        }
      ],
      "evaluation": "Task configuration selects an evaluator and expected application or system outcome; evaluators return a reward for the resulting state.",
      "evidence_urls": [
        "https://raw.githubusercontent.com/microsoft/WindowsAgentArena/HEAD/README.md",
        "https://raw.githubusercontent.com/microsoft/WindowsAgentArena/HEAD/docs/Develop-Tasks.md",
        "https://raw.githubusercontent.com/microsoft/WindowsAgentArena/HEAD/src/win-arena-container/client/evaluation_examples_windows/test_all.json"
      ],
      "focus": [
        "Windows task completion",
        "desktop workflows"
      ],
      "notes": "Counted unique identifiers in the inspected manifest. Normal and hard modes differ in how much task setup the agent must perform. Record the difficulty and observation/backend configuration with each result; no VM was launched.",
      "observations": [
        "screenshots",
        "accessibility information",
        "screen-element detections"
      ],
      "setting": "interactive"
    }
  },
  "environments": [
    "windows",
    "desktop"
  ],
  "evidence": [
    {
      "method": "editorial_primary_review",
      "retrieved_at": "2026-10-16T06:35:57Z",
      "role": "primary",
      "sha256": "e481abd422a4cc8b819ad84b32684e4af641e18a7c0d277dd608cfd5d3c5e6ad",
      "supports": [
        "name",
        "kind",
        "summary",
        "environments",
        "topics",
        "scope",
        "relations",
        "details"
      ],
      "url": "https://raw.githubusercontent.com/microsoft/WindowsAgentArena/HEAD/README.md"
    },
    {
      "method": "editorial_primary_review",
      "retrieved_at": "2026-10-16T06:35:57Z",
      "role": "primary",
      "sha256": "e481abd422a4cc8b819ad84b32684e4af641e18a7c0d277dd608cfd5d3c5e6ad",
      "supports": [
        "details.benchmark_protocol"
      ],
      "url": "https://raw.githubusercontent.com/microsoft/WindowsAgentArena/HEAD/README.md"
    },
    {
      "method": "editorial_primary_review",
      "retrieved_at": "2026-10-17T02:06:30Z",
      "role": "primary",
      "sha256": "c565dc02ab0049bf9710b87e2760dda8d7c92268ea78f0354952695ab59c0de3",
      "supports": [
        "details.benchmark_protocol"
      ],
      "url": "https://raw.githubusercontent.com/microsoft/WindowsAgentArena/HEAD/docs/Develop-Tasks.md"
    },
    {
      "method": "editorial_primary_review",
      "retrieved_at": "2026-10-17T02:06:30Z",
      "role": "primary",
      "sha256": "902fb24cf0a9e146bf6c29288e74a7254fe40294e49dd47363e0b15d65608fef",
      "supports": [
        "details.benchmark_protocol"
      ],
      "url": "https://raw.githubusercontent.com/microsoft/WindowsAgentArena/HEAD/src/win-arena-container/client/evaluation_examples_windows/test_all.json"
    },
    {
      "method": "license_header",
      "retrieved_at": "2026-10-16T06:41:57Z",
      "role": "primary",
      "sha256": "9906940f61b1f0b533fa7d99baf55178b2808fbe113ea51dfbfad8572ccd5f2b",
      "supports": [
        "licenses.code"
      ],
      "url": "https://raw.githubusercontent.com/microsoft/WindowsAgentArena/HEAD/LICENSE"
    }
  ],
  "id": "github:microsoft/windowsagentarena",
  "identifiers": {},
  "kind": "benchmark",
  "last_verified_at": "2026-10-16T06:35:57Z",
  "licenses": {
    "code": {
      "basis": "repository LICENSE header; not a weights/data license",
      "declared": "MIT",
      "retrieved_at": "2026-10-16T06:41:57Z",
      "url": "https://raw.githubusercontent.com/microsoft/WindowsAgentArena/HEAD/LICENSE"
    },
    "data": null,
    "document": null,
    "weights": null
  },
  "links": [
    {
      "rel": "primary",
      "url": "https://github.com/microsoft/WindowsAgentArena"
    }
  ],
  "name": "Windows Agent Arena",
  "notes": "",
  "published_at": null,
  "relations": [
    {
      "evidence_url": "https://raw.githubusercontent.com/microsoft/WindowsAgentArena/HEAD/README.md",
      "target_id": "arxiv:2409.08264",
      "type": "documented_by"
    }
  ],
  "review_status": "curated",
  "scope": "core",
  "summary": "Windows 11 benchmark for multimodal desktop agents in a reproducible VM that runs locally or in parallel on Azure ML, with a harder mode where agents must set tasks up themselves.",
  "topics": [
    "evaluation"
  ],
  "updated_at": null,
  "url": "https://github.com/microsoft/WindowsAgentArena"
}