{"url":"/sota/question-answering-on-drop-test","task":{"name":"Question Answering","url":"/task/question-answering","note":null},"dataset":{"name":"DROP Test","url":"/dataset/drop"},"category":"Natural Language Processing","categories":["Miscellaneous","Natural Language Processing","Reasoning"],"category_note":null,"description":"Question answering can be segmented into domain-specific tasks like community question answering and knowledge-base question answering. Popular benchmark datasets for evaluation question answering systems include [SQuAD](/dataset/squad), [HotPotQA](/dataset/hotpotqa), [bAbI](/dataset/babi-1), [TriviaQA](/dataset/triviaqa), [WikiQA](/dataset/wikiqa), and many others. Models for question answering are typically evaluated on metrics like EM and F1. Some recent top performing models are T5 and XLNet.\r\n\r\n( Image credit: [SQuAD](https://rajpurkar.github.io/mlx/qa-and-squad/) )","description_from":"task","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","rank":"the archive's row order at snapshot; not re-ranked","rows_end_at":"2025-07-28","rows_withheld_as_spam":0,"metric_values":"the archive's strings, untouched"},"metrics":["F1"],"metric_direction":{"note":"inferred from the metric name only (the archive records no direction); null = not inferred, chart draws points only","by_metric":{"F1":"higher"}},"counts":{"rows":16,"rows_with_code":11,"rows_with_paper_page":16,"rows_dated":16,"rows_using_additional_data":1},"rows":[{"rank_in_archive_order":1,"model":"QDGAT (ensemble)","metrics":{"F1":"88.38"},"uses_additional_data":false,"paper_date":"2020-09-16","paper":"/paper/question-directed-graph-attention-network-for","paper_url":"https://arxiv.org/abs/2009.07448v2","paper_title":"Question Directed Graph Attention Network for Numerical Reasoning over Text","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":2,"model":"POET","metrics":{"F1":"87.6"},"uses_additional_data":true,"paper_date":"2022-01-27","paper":"/paper/reasoning-like-program-executors-1","paper_url":"https://arxiv.org/abs/2201.11473v2","paper_title":"Reasoning Like Program Executors","code":"https://github.com/microsoft/ContextualSP","n_code_links":1,"syntology":null},{"rank_in_archive_order":3,"model":"PaLM 2 (few-shot)","metrics":{"F1":"85.0"},"uses_additional_data":false,"paper_date":"2023-05-17","paper":"/paper/palm-2-technical-report-1","paper_url":"https://arxiv.org/abs/2305.10403v3","paper_title":"PaLM 2 Technical Report","code":"https://github.com/eternityyw/tram-benchmark","n_code_links":1,"syntology":null},{"rank_in_archive_order":4,"model":"BERT+Calculator (ensemble)","metrics":{"F1":"81.78"},"uses_additional_data":false,"paper_date":"2019-08-31","paper":"/paper/giving-bert-a-calculator-finding-operations","paper_url":"https://arxiv.org/abs/1909.00109v2","paper_title":"Giving BERT a Calculator: Finding Operations and Arguments with Reading Comprehension","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":5,"model":"NeRd","metrics":{"F1":"81.71"},"uses_additional_data":false,"paper_date":"2020-05-01","paper":"/paper/neural-symbolic-reader-scalable-integration","paper_url":"https://openreview.net/forum?id=ryxjnREFwH","paper_title":"Neural Symbolic Reader: Scalable Integration of Distributed and Symbolic Representations for Reading Comprehension","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":6,"model":"GPT-4 (few-shot, k=3)","metrics":{"F1":"80.9"},"uses_additional_data":false,"paper_date":"2023-03-15","paper":"/paper/gpt-4-technical-report-1","paper_url":"https://arxiv.org/abs/2303.08774v5","paper_title":"GPT-4 Technical Report","code":"https://github.com/openai/evals","n_code_links":11,"syntology":{"n_ran":2,"n_unverified":3,"n_samples":5,"n_pointer_only_licence":1}},{"rank_in_archive_order":7,"model":"TASE-BERT","metrics":{"F1":"80.7"},"uses_additional_data":false,"paper_date":"2019-09-29","paper":"/paper/tag-based-multi-span-extraction-in-reading","paper_url":"https://arxiv.org/abs/1909.13375v4","paper_title":"A Simple and Effective Model for Answering Multi-span Questions","code":"https://github.com/llamazing/numnet_plus","n_code_links":4,"syntology":null},{"rank_in_archive_order":8,"model":"MTMSN Large","metrics":{"F1":"79.88"},"uses_additional_data":false,"paper_date":"2019-08-15","paper":"/paper/a-multi-type-multi-span-network-for-reading","paper_url":"https://arxiv.org/abs/1908.05514v2","paper_title":"A Multi-Type Multi-Span Network for Reading Comprehension that Requires Discrete Reasoning","code":"https://github.com/huminghao16/MTMSN","n_code_links":1,"syntology":null},{"rank_in_archive_order":9,"model":"GenBERT (+ND+TD)","metrics":{"F1":"72.4"},"uses_additional_data":false,"paper_date":"2020-04-09","paper":"/paper/injecting-numerical-reasoning-skills-into","paper_url":"https://arxiv.org/abs/2004.04487v1","paper_title":"Injecting Numerical Reasoning Skills into Language Models","code":"https://github.com/ag1988/injecting_numeracy","n_code_links":2,"syntology":null},{"rank_in_archive_order":10,"model":"NumNet","metrics":{"F1":"67.97"},"uses_additional_data":false,"paper_date":"2019-10-15","paper":"/paper/numnet-machine-reading-comprehension-with","paper_url":"https://arxiv.org/abs/1910.06701v1","paper_title":"NumNet: Machine Reading Comprehension with Numerical Reasoning","code":"https://github.com/ranqiu92/NumNet","n_code_links":2,"syntology":null},{"rank_in_archive_order":11,"model":"GPT 3.5 (few-shot, k=3)","metrics":{"F1":"64.1"},"uses_additional_data":false,"paper_date":"2023-03-15","paper":"/paper/gpt-4-technical-report-1","paper_url":"https://arxiv.org/abs/2303.08774v5","paper_title":"GPT-4 Technical Report","code":"https://github.com/openai/evals","n_code_links":11,"syntology":{"n_ran":2,"n_unverified":3,"n_samples":5,"n_pointer_only_licence":1}},{"rank_in_archive_order":12,"model":"Orca 2-7B","metrics":{"F1":"60.26"},"uses_additional_data":false,"paper_date":"2023-11-18","paper":"/paper/orca-2-teaching-small-language-models-how-to","paper_url":"https://arxiv.org/abs/2311.11045v2","paper_title":"Orca 2: Teaching Small Language Models How to Reason","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":13,"model":"Orca 2-13B","metrics":{"F1":"57.97"},"uses_additional_data":false,"paper_date":"2023-11-18","paper":"/paper/orca-2-teaching-small-language-models-how-to","paper_url":"https://arxiv.org/abs/2311.11045v2","paper_title":"Orca 2: Teaching Small Language Models How to Reason","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":14,"model":"NAQA Net","metrics":{"F1":"47.01"},"uses_additional_data":false,"paper_date":"2019-03-01","paper":"/paper/drop-a-reading-comprehension-benchmark","paper_url":"http://arxiv.org/abs/1903.00161v2","paper_title":"DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs","code":"https://github.com/allenai/allennlp-reading-comprehension","n_code_links":3,"syntology":null},{"rank_in_archive_order":15,"model":"GPT-3 175B (few-shot, k=32)","metrics":{"F1":"36.5"},"uses_additional_data":false,"paper_date":"2020-05-28","paper":"/paper/language-models-are-few-shot-learners","paper_url":"https://arxiv.org/abs/2005.14165v4","paper_title":"Language Models are Few-Shot Learners","code":"https://github.com/ggml-org/llama.cpp","n_code_links":67,"syntology":{"n_ran":15,"n_unverified":50,"n_samples":65,"n_pointer_only_licence":4}},{"rank_in_archive_order":16,"model":"BERT","metrics":{"F1":"32.7"},"uses_additional_data":false,"paper_date":"2019-03-01","paper":"/paper/drop-a-reading-comprehension-benchmark","paper_url":"http://arxiv.org/abs/1903.00161v2","paper_title":"DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs","code":"https://github.com/allenai/allennlp-reading-comprehension","n_code_links":3,"syntology":null}],"since_archive":{"present":false,"note":"No Syntology-extracted rows are published in this build."},"syntology":{"read_at":"2026-09-24T18:15:14+00:00","claim":"Per row: N of M harvested code samples from that row's paper executed on a synthesized fixture; the other M-N are unverified. Not a reproduction of the row's number; not a correctness claim. n_pointer_only_licence counts samples the site points at rather than redistributes (a licence axis, independent of ran/unverified).","rows_with_graph_line":3,"rows_with_any_sample_ran":3,"distinct_papers_with_graph_line":2,"distinct_papers_with_any_sample_ran":2,"samples_over_distinct_papers":{"n_ran":17,"n_unverified":53,"n_samples":70,"n_pointer_only_licence":5,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":19,"n_unverified":56,"n_samples":75,"n_pointer_only_licence":6,"note":"row-weighted: a paper behind several rows is counted once per row; inflated relative to samples_over_distinct_papers by design, kept for readers summing the per-row syntology blocks"}}}