{"paper":{"title":"DeepResearch Bench: A Comprehensive Benchmark for Deep Research Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"DeepResearch Bench supplies 100 PhD-level tasks across 22 fields plus two evaluation methods that align with human judgment for deep research agents.","cross_cats":["cs.IR"],"primary_cat":"cs.CL","authors_text":"Benfeng Xu, Chiwei Zhu, Mingxuan Du, Xiaorui Wang, Zhendong Mao","submitted_at":"2025-06-13T13:17:32Z","abstract_excerpt":"Deep Research Agents are a prominent category of LLM-based agents. By autonomously orchestrating multistep web exploration, targeted retrieval, and higher-order synthesis, they transform vast amounts of online information into analyst-grade, citation-rich reports--compressing hours of manual desk research into minutes. However, a comprehensive benchmark for systematically evaluating the capabilities of these agents remains absent. To bridge this gap, we present DeepResearch Bench, a benchmark consisting of 100 PhD-level research tasks, each meticulously crafted by domain experts across 22 dist"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We present DeepResearch Bench, a benchmark consisting of 100 PhD-level research tasks... We therefore propose two novel methodologies that achieve strong alignment with human judgment.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The 100 tasks crafted by domain experts across 22 fields are representative of real deep-research challenges and the two proposed evaluation methodologies genuinely align with human judgment without introducing systematic bias or requiring undisclosed tuning.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DeepResearch Bench supplies 100 expert-crafted PhD-level tasks and two human-aligned evaluation frameworks to measure deep research agents on report quality and citation accuracy.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"DeepResearch Bench supplies 100 PhD-level tasks across 22 fields plus two evaluation methods that align with human judgment for deep research agents.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"35fd12d796583bb09f15ce1846605ec6f5a274bc95e3fcba1a953874884c3c27"},"source":{"id":"2506.11763","kind":"arxiv","version":1},"verdict":{"id":"0955f400-a456-4404-9685-39e6f7dedc53","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T08:04:42.628782Z","strongest_claim":"We present DeepResearch Bench, a benchmark consisting of 100 PhD-level research tasks... We therefore propose two novel methodologies that achieve strong alignment with human judgment.","one_line_summary":"DeepResearch Bench supplies 100 expert-crafted PhD-level tasks and two human-aligned evaluation frameworks to measure deep research agents on report quality and citation accuracy.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The 100 tasks crafted by domain experts across 22 fields are representative of real deep-research challenges and the two proposed evaluation methodologies genuinely align with human judgment without introducing systematic bias or requiring undisclosed tuning.","pith_extraction_headline":"DeepResearch Bench supplies 100 PhD-level tasks across 22 fields plus two evaluation methods that align with human judgment for deep research agents."},"references":{"count":66,"sample":[{"doi":"","year":2024,"title":"2408.07055 , archiveprefix =","work_id":"7b17f534-444c-40c9-8c05-ead22c892088","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Mle-bench: Evaluating machine learning agents on machine learning engineering","work_id":"a671e43f-ceab-49e7-adc3-473d802a97ca","ref_index":2,"cited_arxiv_id":"2410.07095","is_internal_anchor":true},{"doi":"","year":2025,"title":"ScienceAgentBench: Toward rigorous assessment of language agents for data-driven scientific discovery","work_id":"e1160394-4fe6-4475-9753-e69ad39fb09e","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"deepseek-ai/DeepSeek-V3-0324 · Hugging Face, March 2025","work_id":"1a4c634b-d889-4f66-a3ba-0c5b087bfefa","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning","work_id":"e6b75ad5-2877-4168-97c8-710407094d20","ref_index":5,"cited_arxiv_id":"2501.12948","is_internal_anchor":true}],"resolved_work":66,"snapshot_sha256":"5dcc6e48560925ff898e2a401bf2b79a5a41525f45dcff962f90abe25c56cb88","internal_anchors":19},"formal_canon":{"evidence_count":2,"snapshot_sha256":"f39c9ba62a02cfd3626ee4571502ea64ac1f310e7755d9f61c63408e27581b27"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}