{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2609.12475", "kind": "paper", "url": "https://arxiv.org/abs/2609.12475"}, {"id": "gh:milkthink-lab/zipbench", "kind": "repo", "url": "https://github.com/MilkThink-Lab/ZipBench"}], "authors": ["Zhongzhan Huang", "Junxin Li", "Guoming Ling", "Yupei Lin", "Shanshan Zhong", "Hefeng Wu"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2609.12475", "source_url": "https://arxiv.org/abs/2609.12475"}, "categories": ["General AI", "Language & Knowledge", "cs.CL"], "description": {"en": "A collection of compressed benchmark versions for text, multimodal, and agent tasks, with tools to evaluate models or create new compact benchmarks."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2609.12475", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2609.12475", "title": "ZipBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2609.12475", "modality": null, "name": "ZipBench", "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": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_zipbench_638d2f5b", "origin_source": "arxiv", "recognition_confidence": "0.75", "record_sha256": "df8404f756632fecfbf822ea8238151af08b4c01d3cab5a061726807cd0ad437", "relation": "introduces", "review_model": "deepseek-v4-pro", "review_state": "ai-reviewed", "reviewed_at": "2026-09-17T09:37:12.088376Z", "source_url": "https://arxiv.org/abs/2609.12475"}, "publisher": null, "released": "2026-09-11", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2609.12475", "source_url": "https://arxiv.org/abs/2609.12475"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2609-12475", "source": "claire_radar", "source_benchmark_id": "2609.12475", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attention": {"asOf": "2026-09-14", "attemptedAt": "2026-09-14T01:54:58.773663Z", "githubRepo": "https://github.com/MilkThink-Lab/ZipBench", "githubScope": "benchmark_repo", "githubStars": 4, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": null, "hfPaperUrl": "https://huggingface.co/papers/2609.12475", "observedAt": "2026-09-14T01:54:58.773663Z", "signalStatus": {"githubStars": {"observedAt": "2026-09-14T01:54:58.773663Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-09-14T01:54:58.773663Z", "state": "unavailable"}}, "status": "fresh"}, "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "EMNLP acceptance, broad coverage of popular benchmarks, and clear cost-saving value position it for rapid adoption and discussion in NLP evaluation communities.", "score": 90}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-17T09:37:12.088376Z", "inputHash": "ea2896d2e835ffc94a6a5413719581ec63881fe5636e9e5fa829e91cdd213559", "model": "deepseek-v4-pro"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "ZipBench: Low-Cost Framework for Compressing Comprehensive Benchmarks of Large Language Models", "canonicalNameSource": "paper_title", "decisionReason": "The repository provides ready-to-run evaluation scripts for multiple benchmark families and a well-defined methodology, enabling external use and extension.", "evaluationEvidence": {"reuse": "Evaluate a text LLM: cd opencompass && python run.py zipbench/configs/eval_arc_challenge_small.py; multimodal: --subset small; agents: zipbench-agent list and score", "scoring": "Weighted estimates of full-benchmark scores for small and tiny versions, with fidelity measured by MAE (0.008 for small, 0.015 for tiny) and Spearman correlation (0.99 and 0.98 respectively)", "task": "Compact versions of 100+ text, multimodal, and agent benchmarks, reducing evaluation cost by 60-80% while preserving full-benchmark scores and rankings"}, "model": "deepseek-v4-pro", "reviewedAt": "2026-09-17T09:37:12.088376Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A collection of compressed benchmark versions for text, multimodal, and agent tasks, with tools to evaluate models or create new compact benchmarks.", "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": "To address this challenge, we present ZipBench, a simple and low-cost BCM with theoretical error and rank-consistency guarantees."}, "familyId": "bmf_71f839953810", "firstSeenAt": "2026-09-14", "heat": null, "id": "bm_zipbench_638d2f5b", "indexedAt": "2026-09-14T00:39:56.575498Z", "industrySectors": [], "links": {"code": "https://github.com/MilkThink-Lab/ZipBench", "data": null, "paper": "https://arxiv.org/abs/2609.12475", "pdf": "https://arxiv.org/pdf/2609.12475", "project": null, "report": "https://arxiv.org/abs/2609.12475"}, "metrics": [], "motivation": "Comprehensive benchmark suites are essential for improving large language models (LLMs), but many widely used benchmarks are redundant, making evaluation unnecessarily expensive.", "name": "ZipBench", "oneLine": "A collection of compressed benchmark versions for text, multimodal, and agent tasks, with tools to evaluate models or create new compact benchmarks.", "paperTitle": "Zipbench: Low-Cost Framework for Compressing Comprehensive Benchmarks of Large Language Models", "primaryDomain": "General AI", "publication": {"evidence": "Accepted by EMNLP 2026 main track", "evidenceLevel": "author-claim", "evidenceUrl": "https://arxiv.org/abs/2609.12475", "source": "arxiv-comments", "status": "acceptance_claimed", "venue": "EMNLP 2026 main track", "verifiedAt": "2026-09-14T00:39:56.575498Z"}, "publications": [], "ranking": {"30d": {"confidence": "Low", "coverage": 0.55, "growth": {"components": {"githubStars": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 30}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": 0.2397076310863079, "value": 4}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": null, "value": null}}, "confidence": "Low", "coverage": 0.55, "rank": 72, "score": 36}, "method": "attention-ranking-v13", "rank": 72, "score": 36}, "90d": {"confidence": "Low", "coverage": 0.55, "growth": {"components": {"githubStars": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 90}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": 0.21659283491533646, "value": 4}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": null, "value": null}}, "confidence": "Low", "coverage": 0.55, "rank": 196, "score": 34}, "method": "attention-ranking-v13", "rank": 196, "score": 34}}, "readiness": "Runnable", "recognitionConfidence": 0.75, "relation": "introduces", "releasedAt": "2026-09-11", "source": {"authors": ["Zhongzhan Huang", "Junxin Li", "Guoming Ling", "Yupei Lin", "Shanshan Zhong", "Hefeng Wu"], "categories": ["cs.CL"], "id": "2609.12475", "title": "Zipbench: Low-Cost Framework for Compressing Comprehensive Benchmarks of Large Language Models", "type": "arxiv", "url": "https://arxiv.org/abs/2609.12475"}, "sourceUpdatedAt": "2026-09-11T06:15:25Z", "topics": ["cs.CL"], "venueAttempts": [{"decisionRaw": "Accepted by EMNLP 2026 main track", "evidence": [{"level": "author-claim", "observedAt": "2026-09-14T00:39:56.575498Z", "rawValue": "Accepted by EMNLP 2026 main track", "sourceType": "arxiv-comments", "sourceUrl": "https://arxiv.org/abs/2609.12475"}], "reviewStatus": "accepted", "venueName": "EMNLP 2026 main track"}], "whyItMatters": "Reduces evaluation costs dramatically while maintaining ranking fidelity, making broad model assessment feasible for researchers with limited compute."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
