{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.16748", "kind": "paper", "url": "https://arxiv.org/abs/2606.16748"}, {"kind": "website", "url": "https://mypcbench.com"}], "authors": ["Lawrence Keunho Jang", "Andrew Keunwoo Jang", "Jing Yu Koh", "Ruslan Salakhutdinov"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.16748", "source_url": "https://arxiv.org/abs/2606.16748"}, "categories": ["General AI", "Language & Knowledge", "cs.LG"], "description": {"en": "MyPCBench tests computer-use agents as personal assistants on a Linux desktop seeded with 17 simulated web apps for a single persona. It has 184 tasks."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.16748", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.16748", "title": "MyPCBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Carnegie Mellon University"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.16748", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.16748", "source_url": "https://openalex.org/W7164880355"}, "key": "claire-radar:2606.16748", "modality": null, "name": "MyPCBench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-09-25T04:40:22Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_mypcbench_7c7e41a1", "origin_source": "arxiv", "recognition_confidence": "0.75", "record_sha256": "9f8f5e145c809a416e60db748ecdd019149e7cae449279c3e263721b32d906c6", "relation": "introduces", "review_model": "claude-code-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T16:31:14.356495Z", "source_url": "https://arxiv.org/abs/2606.16748"}, "publisher": null, "released": "2026-06-15", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.16748", "source_url": "https://arxiv.org/abs/2606.16748"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-16748", "source": "claire_radar", "source_benchmark_id": "2606.16748", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "Computer-use agent benchmarks are drawing active interest, and this release has a project site and results for frontier models.", "score": 55}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T16:31:14.356495Z", "inputHash": "0118463c784b6bcb272d61ecb2a8e631af77d670d957b9d38d87116a088e7d2c", "model": "claude-code-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "MyPCBench: A Benchmark for Personally Intelligent Computer-Use Agents", "canonicalNameSource": "paper_title", "decisionReason": "The benchmark is formally named, with a fixed set of 184 tasks, a uniform harness, a task-solve-rate metric that allows comparison across models, and a publicly released environment and harness.", "evaluationEvidence": {"reuse": "We release the environment, task set, and agent harness", "scoring": "the best model, Claude Opus 4.6, fully solves", "task": "We define 184 tasks in this environment"}, "model": "claude-code-default", "reviewedAt": "2026-09-24T16:31:14.356495Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "MyPCBench tests computer-use agents as personal assistants on a Linux desktop seeded with 17 simulated web apps for a single persona. It has 184 tasks.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "evaluation protocol evidence", "public artifact URL"], "snippet": "We introduce MyPCBench, which tests computer-use agents as personal assistants on a Linux desktop populated with 17 simulated real-world web applications and a full desktop stack, all seeded for one canonical persona, Michael Scott from The Office."}, "familyId": "bmf_e7f42ac35e0c", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_mypcbench_7c7e41a1", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "paper": "https://arxiv.org/abs/2606.16748", "pdf": "https://arxiv.org/pdf/2606.16748", "project": "https://mypcbench.com", "report": "https://arxiv.org/abs/2606.16748"}, "metrics": [], "motivation": "Current benchmarks for computer-use agents evaluate models in impersonal environments.", "name": "MyPCBench", "oneLine": "MyPCBench tests computer-use agents as personal assistants on a Linux desktop seeded with 17 simulated web apps for a single persona. It has 184 tasks.", "paperTitle": "MyPCBench: A Benchmark for Personally Intelligent Computer-Use Agents", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.16748", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "readiness": "Paper only", "recognitionConfidence": 0.75, "relation": "introduces", "releasedAt": "2026-06-15", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Lawrence Keunho Jang", "Andrew Keunwoo Jang", "Jing Yu Koh", "Ruslan Salakhutdinov"], "categories": ["cs.LG", "cs.CL"], "id": "2606.16748", "title": "MyPCBench: A Benchmark for Personally Intelligent Computer-Use Agents", "type": "arxiv", "url": "https://arxiv.org/abs/2606.16748"}, "sourceUpdatedAt": "2026-06-15T14:08:09Z", "topics": ["cs.LG"], "venueAttempts": [], "whyItMatters": "It covers personalized, logged-in computer-use tasks that live web benchmarks cannot run, which narrows the gap between how assistants are evaluated and how they are deployed."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
