{"record": {"aliases": ["ClawBench", "ClawBench: Can AI Agents Complete Everyday Online Tasks?"], "artifacts": [{"kind": "paper", "url": "https://arxiv.org/abs/2604.08523"}], "audits": [], "categories": ["agent"], "description": {"en": "Agents complete everyday tasks on live production websites. The paper evaluates 153 tasks with trajectory judging against human references; the current repository lists a changed 152-task V1 corpus. V2 uses request interception and separate lenient and strict judge rubrics. Corpus revision, harness and judge change the measurement, and live websites can change between runs."}, "documents": [{"document_type": "benchmark_leaderboard", "id": "model_reports:clawbench_v1_paper_results", "model_name": null, "organization": "TIGER-AI-Lab", "published": "2026-07-20", "retrieved_at": "2026-11-12", "source": "model_reports", "source_id": "clawbench_v1_paper_results", "source_url": "https://arxiv.org/html/2604.08523v2#S3.T3", "title": "ClawBench paper Table 3 (arXiv v2)"}], "evidence_summary": {"document_count": 1, "model_count": 8, "model_count_basis": "source_model_id"}, "key": "model-reports:clawbench", "modality": null, "name": "ClawBench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"registry": "data/model_cards.yml", "score_archive": "data/benchmark_scores.yml", "source_url": "https://arxiv.org/abs/2604.08523"}, "publisher": null, "released": "2026-04-09", "released_reference": {"basis": "benchmark_release", "source_key": "model-reports:clawbench", "source_url": "https://arxiv.org/abs/2604.08523"}, "schema_version": 1, "sizes": [], "slug": "clawbench", "source": "model_reports", "source_benchmark_id": "clawbench"}, "schema_version": 1, "scores_by_source": {"model_reports": {"rows": [{"benchmark_id": "clawbench", "comparable_group": "[\"model-reports:clawbench\",\"clawbench_paper_153_tasks\",\"OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\"]", "date_precision": "document_publication", "document_id": "model_reports:clawbench_v1_paper_results", "instrument": "clawbench_paper_153_tasks", "key": "model-reports:clawbench", "measured_by": "TIGER-AI-Lab", "measurement_kind": "benchmark_publisher_run", "model": "Claude Haiku 4.5", "model_id": "[\"Anthropic\",\"Claude Haiku 4.5\"]", "model_name": "Claude Haiku 4.5", "obs_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Claude Haiku 4.5", "observation_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Claude Haiku 4.5", "organization": "Anthropic", "protocol": "OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout", "rank_in_source_response": null, "raw_value": "18.3", "read_from": "html_text", "reported_at": "2026-07-20", "reported_by": "third_party", "reported_date": "2026-07-20", "series_id": "model_reports:clawbench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "clawbench_v1_paper_results", "source_title": "ClawBench paper Table 3 (arXiv v2)", "source_url": "https://arxiv.org/html/2604.08523v2#S3.T3", "value": 18.3, "value_kind": "number"}, {"benchmark_id": "clawbench", "comparable_group": "[\"model-reports:clawbench\",\"clawbench_paper_153_tasks\",\"OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\"]", "date_precision": "document_publication", "document_id": "model_reports:clawbench_v1_paper_results", "instrument": "clawbench_paper_153_tasks", "key": "model-reports:clawbench", "measured_by": "TIGER-AI-Lab", "measurement_kind": "benchmark_publisher_run", "model": "GLM-5", "model_id": "[\"Z.ai\",\"GLM-5\"]", "model_name": "GLM-5", "obs_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000GLM-5", "observation_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000GLM-5", "organization": "Z.ai", "protocol": "OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout", "rank_in_source_response": null, "raw_value": "24.2", "read_from": "html_text", "reported_at": "2026-07-20", "reported_by": "third_party", "reported_date": "2026-07-20", "series_id": "model_reports:clawbench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "clawbench_v1_paper_results", "source_title": "ClawBench paper Table 3 (arXiv v2)", "source_url": "https://arxiv.org/html/2604.08523v2#S3.T3", "value": 24.2, "value_kind": "number"}, {"benchmark_id": "clawbench", "comparable_group": "[\"model-reports:clawbench\",\"clawbench_paper_153_tasks\",\"OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\"]", "date_precision": "document_publication", "document_id": "model_reports:clawbench_v1_paper_results", "instrument": "clawbench_paper_153_tasks", "key": "model-reports:clawbench", "measured_by": "TIGER-AI-Lab", "measurement_kind": "benchmark_publisher_run", "model": "Gemini 3 Flash", "model_id": "[\"Google\",\"Gemini 3 Flash\"]", "model_name": "Gemini 3 Flash", "obs_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Gemini 3 Flash", "observation_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Gemini 3 Flash", "organization": "Google", "protocol": "OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout", "rank_in_source_response": null, "raw_value": "19.0", "read_from": "html_text", "reported_at": "2026-07-20", "reported_by": "third_party", "reported_date": "2026-07-20", "series_id": "model_reports:clawbench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "clawbench_v1_paper_results", "source_title": "ClawBench paper Table 3 (arXiv v2)", "source_url": "https://arxiv.org/html/2604.08523v2#S3.T3", "value": 19.0, "value_kind": "number"}, {"benchmark_id": "clawbench", "comparable_group": "[\"model-reports:clawbench\",\"clawbench_paper_153_tasks\",\"OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\"]", "date_precision": "document_publication", "document_id": "model_reports:clawbench_v1_paper_results", "instrument": "clawbench_paper_153_tasks", "key": "model-reports:clawbench", "measured_by": "TIGER-AI-Lab", "measurement_kind": "benchmark_publisher_run", "model": "Gemini 3.1 Flash-Lite", "model_id": "[\"Google\",\"Gemini 3.1 Flash-Lite\"]", "model_name": "Gemini 3.1 Flash-Lite", "obs_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Gemini 3.1 Flash-Lite", "observation_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Gemini 3.1 Flash-Lite", "organization": "Google", "protocol": "OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout", "rank_in_source_response": null, "raw_value": "3.3", "read_from": "html_text", "reported_at": "2026-07-20", "reported_by": "third_party", "reported_date": "2026-07-20", "series_id": "model_reports:clawbench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "clawbench_v1_paper_results", "source_title": "ClawBench paper Table 3 (arXiv v2)", "source_url": "https://arxiv.org/html/2604.08523v2#S3.T3", "value": 3.3, "value_kind": "number"}, {"benchmark_id": "clawbench", "comparable_group": "[\"model-reports:clawbench\",\"clawbench_paper_153_tasks\",\"OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\"]", "date_precision": "document_publication", "document_id": "model_reports:clawbench_v1_paper_results", "instrument": "clawbench_paper_153_tasks", "key": "model-reports:clawbench", "measured_by": "TIGER-AI-Lab", "measurement_kind": "benchmark_publisher_run", "model": "Gemini 3.1 Pro", "model_id": "[\"Google\",\"Gemini 3.1 Pro\"]", "model_name": "Gemini 3.1 Pro", "obs_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Gemini 3.1 Pro", "observation_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Gemini 3.1 Pro", "organization": "Google", "protocol": "OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout", "rank_in_source_response": null, "raw_value": "9.8", "read_from": "html_text", "reported_at": "2026-07-20", "reported_by": "third_party", "reported_date": "2026-07-20", "series_id": "model_reports:clawbench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "clawbench_v1_paper_results", "source_title": "ClawBench paper Table 3 (arXiv v2)", "source_url": "https://arxiv.org/html/2604.08523v2#S3.T3", "value": 9.8, "value_kind": "number"}, {"benchmark_id": "clawbench", "comparable_group": "[\"model-reports:clawbench\",\"clawbench_paper_153_tasks\",\"OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\"]", "date_precision": "document_publication", "document_id": "model_reports:clawbench_v1_paper_results", "instrument": "clawbench_paper_153_tasks", "key": "model-reports:clawbench", "measured_by": "TIGER-AI-Lab", "measurement_kind": "benchmark_publisher_run", "model": "Qwen3.5-397B-A17B", "model_id": "[\"Qwen\",\"Qwen3.5-397B-A17B\"]", "model_name": "Qwen3.5-397B-A17B", "obs_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Qwen3.5-397B-A17B", "observation_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Qwen3.5-397B-A17B", "organization": "Qwen", "protocol": "OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout", "rank_in_source_response": null, "raw_value": "26.1", "read_from": "html_text", "reported_at": "2026-07-20", "reported_by": "third_party", "reported_date": "2026-07-20", "series_id": "model_reports:clawbench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "clawbench_v1_paper_results", "source_title": "ClawBench paper Table 3 (arXiv v2)", "source_url": "https://arxiv.org/html/2604.08523v2#S3.T3", "value": 26.1, "value_kind": "number"}, {"benchmark_id": "clawbench", "comparable_group": "[\"model-reports:clawbench\",\"clawbench_paper_153_tasks\",\"OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; high reasoning effort; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\"]", "date_precision": "document_publication", "document_id": "model_reports:clawbench_v1_paper_results", "instrument": "clawbench_paper_153_tasks", "key": "model-reports:clawbench", "measured_by": "TIGER-AI-Lab", "measurement_kind": "benchmark_publisher_run", "model": "Claude Sonnet 4.6", "model_id": "[\"Anthropic\",\"Claude Sonnet 4.6\"]", "model_name": "Claude Sonnet 4.6", "obs_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; high reasoning effort; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Claude Sonnet 4.6", "observation_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; high reasoning effort; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Claude Sonnet 4.6", "organization": "Anthropic", "protocol": "OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; high reasoning effort; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout", "rank_in_source_response": null, "raw_value": "33.3", "read_from": "html_text", "reported_at": "2026-07-20", "reported_by": "third_party", "reported_date": "2026-07-20", "series_id": "model_reports:clawbench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "clawbench_v1_paper_results", "source_title": "ClawBench paper Table 3 (arXiv v2)", "source_url": "https://arxiv.org/html/2604.08523v2#S3.T3", "value": 33.3, "value_kind": "number"}, {"benchmark_id": "clawbench", "comparable_group": "[\"model-reports:clawbench\",\"clawbench_paper_153_tasks\",\"OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; high reasoning effort; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\"]", "date_precision": "document_publication", "document_id": "model_reports:clawbench_v1_paper_results", "instrument": "clawbench_paper_153_tasks", "key": "model-reports:clawbench", "measured_by": "TIGER-AI-Lab", "measurement_kind": "benchmark_publisher_run", "model": "GPT-5.4", "model_id": "[\"OpenAI\",\"GPT-5.4\"]", "model_name": "GPT-5.4", "obs_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; high reasoning effort; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000GPT-5.4", "observation_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; high reasoning effort; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000GPT-5.4", "organization": "OpenAI", "protocol": "OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; high reasoning effort; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout", "rank_in_source_response": null, "raw_value": "6.5", "read_from": "html_text", "reported_at": "2026-07-20", "reported_by": "third_party", "reported_date": "2026-07-20", "series_id": "model_reports:clawbench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "clawbench_v1_paper_results", "source_title": "ClawBench paper Table 3 (arXiv v2)", "source_url": "https://arxiv.org/html/2604.08523v2#S3.T3", "value": 6.5, "value_kind": "number"}], "series": {"declared_max": null, "direction": "higher_is_better", "direction_basis": "reported_metric", "display_scale": null, "first_score_record": {"date_precision": "document_publication", "obs_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Claude Haiku 4.5", "reported_at": "2026-07-20", "source_url": "https://arxiv.org/html/2604.08523v2#S3.T3"}, "first_score_report": {"date_precision": "document_publication", "obs_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; greedy decoding at provider default; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Claude Haiku 4.5", "reported_at": "2026-07-20", "source_url": "https://arxiv.org/html/2604.08523v2#S3.T3"}, "key": "model-reports:clawbench", "max_score_contradicted": false, "metric": "success_rate", "observation_count": 8, "observed_max": 33.3, "score_summary": {"display_max": 33.3, "display_multiplier": 1, "model_count": 8, "model_count_basis": "source_model_id", "numeric_count": 8, "raw_max": 33.3, "raw_min": 3.3, "source_reference": {"obs_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; high reasoning effort; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Claude Sonnet 4.6", "observation_id": "curated\u0000clawbench\u0000clawbench_v1_paper_results\u0000clawbench_paper_153_tasks\u0000OpenClaw harness; 153 live-web tasks; trajectory Agent-as-Judge (Claude Sonnet 4.6) against human references; paper Table 3 overall success rate; high reasoning effort; provider-default top_p (1.0); 8192 output tokens per turn; 1800s wall-clock limit; 5min idle timeout\u0000Claude Sonnet 4.6", "reported_at": "2026-07-20", "reported_date": "2026-07-20", "source_id": "clawbench_v1_paper_results", "source_url": "https://arxiv.org/html/2604.08523v2#S3.T3"}, "unit": "percent"}, "series_id": "model_reports:clawbench:default", "unit": "percent"}}}, "siblings": []}