{"paper":{"title":"SEED-Bench: Benchmarking Multimodal LLMs with Generative Comprehension","license":"http://creativecommons.org/licenses/by/4.0/","headline":"SEED-Bench supplies 19K human-verified multiple-choice questions to measure multimodal LLMs on image and video comprehension across 12 dimensions.","cross_cats":["cs.CV"],"primary_cat":"cs.CL","authors_text":"Bohao Li, Guangzhi Wang, Rui Wang, Ying Shan, Yixiao Ge, Yuying Ge","submitted_at":"2023-07-30T04:25:16Z","abstract_excerpt":"Based on powerful Large Language Models (LLMs), recent generative Multimodal Large Language Models (MLLMs) have gained prominence as a pivotal research area, exhibiting remarkable capability for both comprehension and generation. In this work, we address the evaluation of generative comprehension in MLLMs as a preliminary step towards a comprehensive assessment of generative models, by introducing a benchmark named SEED-Bench. SEED-Bench consists of 19K multiple choice questions with accurate human annotations (x 6 larger than existing benchmarks), which spans 12 evaluation dimensions includin"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"SEED-Bench consists of 19K multiple choice questions with accurate human annotations (x 6 larger than existing benchmarks), which spans 12 evaluation dimensions including the comprehension of both the image and video modality.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the pipeline of automatic question generation plus manual verification produces questions that genuinely measure generative comprehension rather than artifacts of the generation process itself.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SEED-Bench is a new benchmark of 19K multiple-choice questions for evaluating generative comprehension in multimodal LLMs across 12 image and video dimensions.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"SEED-Bench supplies 19K human-verified multiple-choice questions to measure multimodal LLMs on image and video comprehension across 12 dimensions.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"446d3742bfce8b2df877ed41886b90a9833d90d6939bcd7a74319525b99f20e2"},"source":{"id":"2307.16125","kind":"arxiv","version":2},"verdict":{"id":"dede37a3-26a2-4e9a-870a-4cc4cef5795b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T16:55:12.675278Z","strongest_claim":"SEED-Bench consists of 19K multiple choice questions with accurate human annotations (x 6 larger than existing benchmarks), which spans 12 evaluation dimensions including the comprehension of both the image and video modality.","one_line_summary":"SEED-Bench is a new benchmark of 19K multiple-choice questions for evaluating generative comprehension in multimodal LLMs across 12 image and video dimensions.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the pipeline of automatic question generation plus manual verification produces questions that genuinely measure generative comprehension rather than artifacts of the generation process itself.","pith_extraction_headline":"SEED-Bench supplies 19K human-verified multiple-choice questions to measure multimodal LLMs on image and video comprehension across 12 dimensions."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2307.16125/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":42,"sample":[{"doi":"","year":2022,"title":"Scaling Instruction-Finetuned Language Models","work_id":"8405abb1-7558-4fdf-af24-f4c52fa77a06","ref_index":1,"cited_arxiv_id":"2210.11416","is_internal_anchor":true},{"doi":"","year":2023,"title":"Gpt-4 technical report","work_id":"388f534c-855a-4366-b933-f07bf3e2db5f","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Introducing chatgpt","work_id":"1b840f0b-fb6b-43d7-bc36-05b0a98a171e","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"FastChat. Vicuna. https://github.com/lm-sys/FastChat, 2023","work_id":"ee2e3957-facb-454c-b74c-07fdc354f73e","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"LLaMA: Open and Efficient Foundation Language Models","work_id":"c018fc23-6f3f-4035-9d02-28a2173b2b9d","ref_index":5,"cited_arxiv_id":"2302.13971","is_internal_anchor":true}],"resolved_work":42,"snapshot_sha256":"b967801c3b760b0eff858d80d9e18d6823f2cfd7743a7f5011bd81be2bf3f7e3","internal_anchors":16},"formal_canon":{"evidence_count":2,"snapshot_sha256":"488a2e33743f22e32483cda8e09b1b26e774339717002e8ebfc44bb5f09d6748"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}