Browse › Natural Language Processing › Question Answering › COPA
COPA Benchmark (Question Answering)
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, HotPotQA, bAbI, TriviaQA, 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.
The archive carries no text for this table; the description above is the archive's text for the task Question Answering. archive 2025-07-28
Over time archive 2025-07-28
The chart needs JavaScript; the table below carries every value.
Direction inferred from the metric name, not from the archive: Accuracy (higher is better). Points are placed at the row's paper date; 60 of 60 rows carry one.
Results archive 2025-07-28
Archive rows end at the archive snapshot, 2025-07-28: no result published after that date is in this table. Rank is the archive's row order at that snapshot; not re-ranked here. Metric values are the archive's strings. Column headers sort the table in your browser; each row keeps its archive rank.
| Paper | Code | Ran Syntology | Report | |||||
|---|---|---|---|---|---|---|---|---|
| 1 | PaLM 540B (finetuned) | 100 | – | Paper | Code | 2022 | 30 of 37 ran · 7 unverified | report |
| 2 | Vega v2 6B (KD-based prompt transfer) | 99.4 | – | Paper | – | 2022 | no code linked | report |
| 3 | ST-MoE-32B 269B (fine-tuned) | 99.2 | – | Paper | Code | 2022 | 5 of 5 ran · 0 unverified | report |
| 4 | UL2 20B (fine-tuned) | 99 | – | Paper | Code | 2022 | 0 of 16 ran · 16 unverified | report |
| 5 | DeBERTa-Ensemble | 98.4 | – | Paper | Code | 2020 | 4 of 13 ran · 9 unverified | report |
| 6 | Turing NLR v5 XXL 5.4B (fine-tuned) | 98.2 | – | Paper | – | 2022 | no code linked | report |
| 7 | DeBERTa-1.5B | 96.8 | – | Paper | Code | 2020 | 4 of 13 ran · 9 unverified | report |
| 8 | PaLM 2-L (1-shot) | 96.0 | – | Paper | Code | 2023 | linked, not harvested | report |
| 9 | T5-XXL 11B (fine-tuned) | 94.8 | – | Paper | Code | 2019 | 2 of 31 ran · 29 unverified | report |
| 10 | FLAN 137B (prompt-tuned) | 94 | – | Paper | Code | 2021 | 0 of 1 ran · 1 unverified | report |
| 11 | GPT-3 175B (few-shot, k=32) | 92 | – | Paper | Code | 2020 | 15 of 65 ran · 50 unverified | report |
| 12 | T5-XL 3B (fine-tuned) | 92 | – | Paper | Code | 2019 | 2 of 31 ran · 29 unverified | report |
| 13 | FLAN 137B (zero-shot) | 91 | – | Paper | Code | 2021 | 0 of 1 ran · 1 unverified | report |
| 14 | ST-MoE-L 4.1B (fine-tuned) | 91 | – | Paper | Code | 2022 | 5 of 5 ran · 0 unverified | report |
| 15 | GPT-3 175B (0-shot) | 91 | – | Paper | Code | 2020 | 15 of 65 ran · 50 unverified | report |
| 16 | T0-3B (CoT fine-tuned) | 90.9 | – | Paper | Code | 2023 | linked, not harvested | report |
| 17 | RoBERTa-Winogrande-ft 355M (fine-tuned) | 90.6 | – | Paper | Code | 2019 | linked, not harvested | report |
| 18 | PaLM 2-M (1-shot) | 90.0 | – | Paper | Code | 2023 | linked, not harvested | report |
| 19 | Flipped-3B | 89.88 | – | Paper | Code | 2022 | linked, not harvested | report |
| 20 | PaLM 2-S (1-shot) | 89.0 | – | Paper | Code | 2023 | linked, not harvested | report |
| 21 | GPT-NeoX (one-shot) | 88 | – | Paper | Code | 2023 | linked, not harvested | report |
| 22 | FLAN 137B (few-shot, k=16) | 87 | – | Paper | Code | 2021 | 0 of 1 ran · 1 unverified | report |
| 23 | GPT-3 175B (1-shot) | 87 | – | Paper | Code | 2020 | 15 of 65 ran · 50 unverified | report |
| 24 | RoBERTa-ft 355M (fine-tuned) | 86.4 | – | Paper | Code | 2019 | linked, not harvested | report |
| 25 | Bloomberg GPT (one-shot) | 86 | – | Paper | Code | 2023 | linked, not harvested | report |
| 26 | OPT 66B (one-shot) | 86 | – | Paper | Code | 2023 | linked, not harvested | report |
| 27 | GPT-3 13B (few-shot, k=32) | 86 | – | Paper | Code | 2020 | 15 of 65 ran · 50 unverified | report |
| 28 | KiC-770M | 85.30 | – | Paper | – | 2022 | no code linked | report |
| 29 | UL2 20B (0-shot) | 85 | – | Paper | Code | 2022 | 0 of 16 ran · 16 unverified | report |
| 30 | RoBERTa-Winogrande 355M (fine-tuned) | 84.4 | – | Paper | Code | 2019 | linked, not harvested | report |
| 31 | Neo-6B (QA + WS) | 84.0 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 32 | BLOOM 176B (one-shot) | 84 | – | Paper | Code | 2023 | linked, not harvested | report |
| 33 | T5-Large 770M (fine-tuned) | 83.4 | – | Paper | Code | 2019 | 2 of 31 ran · 29 unverified | report |
| 34 | BERT-SocialIQA 340M | 83.4 | – | Paper | Code | 2019 | 0 of 10 ran · 10 unverified | report |
| 35 | Hybrid H3 2.7B (0-shot, logit scoring) | 81 | – | Paper | Code | 2022 | 7 of 15 ran · 8 unverified | report |
| 36 | BERT-large 340M | 80.8 | – | Paper | Code | 2019 | 0 of 10 ran · 10 unverified | report |
| 37 | RoE-3B | 79.25 | – | Paper | Code | 2023 | 3 of 3 ran · 0 unverified | report |
| 38 | sMLP – deterministic 9.4B (0-shot) | 79 | – | Paper | – | 2022 | no code linked | report |
| 39 | KELM (finetuning BERT-large based single model) | 78.0 | – | Paper | Code | 2021 | 1 of 6 ran · 5 unverified | report |
| 40 | AlexaTM 20B | 78.0 | – | Paper | Code | 2022 | 1 of 1 ran · 0 unverified | report |
| 41 | Neo-6B (few-shot) | 77.0 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 42 | Hybrid H3 2.7B (3-shot, logit scoring) | 77 | – | Paper | Code | 2022 | 7 of 15 ran · 8 unverified | report |
| 43 | Causal Strength w/multi-word predicates (presumably on WinoGrande?) | 76.4 | – | Paper | Code | 2019 | linked, not harvested | report |
| 44 | Gshard 9B | 76 | – | Paper | – | 2022 | no code linked | report |
| 45 | Switch Transformer 9B | 75 | – | Paper | – | 2022 | no code linked | report |
| 46 | GPT-3 Large 760M (0-shot) | 73.0 | – | Paper | Code | 2020 | 15 of 65 ran · 50 unverified | report |
| 47 | Causal Strength Computation w/multi-word predicates (on ClueWeb12) | 71.2 | – | Paper | – | 2017 | no code linked | report |
| 48 | T5-Base 220M (fine-tuned) | 71.2 | – | Paper | Code | 2019 | 2 of 31 ran · 29 unverified | report |
| 49 | Causal Strength Computation (on Causal Net) | 70.2 | – | Paper | – | 2017 | no code linked | report |
| 50 | Causal Strength Computation (on ClueWeb12) | 69.9 | – | Paper | – | 2017 | no code linked | report |
| 51 | Hybrid H3 125M (0-shot, logit scoring) | 67 | – | Paper | Code | 2022 | 7 of 15 ran · 8 unverified | report |
| 52 | Hybrid H3 125M (0-shot, rank classification) | 67 | – | Paper | Code | 2022 | 7 of 15 ran · 8 unverified | report |
| 53 | Pointwise Mutual Information (on 10M stories) | 65.4 | – | Paper | Code | 2019 | linked, not harvested | report |
| 54 | HASH Layers 10B (0-shot) | 64 | – | Paper | – | 2022 | no code linked | report |
| 55 | Base Layers 10B (0-shot) | 63 | – | Paper | – | 2022 | no code linked | report |
| 56 | N-Grammer 343M | 60.0 | – | Paper | Code | 2022 | 0 of 6 ran · 6 unverified | report |
| 57 | Pointwise Mutual Information (on Project Gutenberg) | 58.8 | – | Paper | – | 2017 | no code linked | report |
| 58 | Neo-6B (QA) | 58.2 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 59 | H3 125M (0-shot, rank classification) | 51 | – | Paper | Code | 2022 | 7 of 15 ran · 8 unverified | report |
| 60 | Random chance baseline | 50 | – | Paper | – | 2017 | no code linked | report |
All 60 rows shown. 60 link to a paper page on this site; 0 are marked as using additional training data in the archive. No GitHub stars are tracked; "Code" is the first repository the archive lists for the row. The archive carries no row tags, review links or community-submitted rows for this table; none are shown. archive 2025-07-28
Syntology Ran reads "N of M ran · U unverified": of the M code samples Syntology harvested from repositories linked to that row's paper (joined by arXiv id), N executed on a synthesized input and the other U = M−N are unverified (harvested, no recorded run). It counts code from repositories linked to that row's paper, not this result: the row's number was not reproduced and nothing here is a correctness claim. The other cell texts mean no graph line for the row: "linked, not harvested" (the archive links code, Syntology has not harvested it), "no code linked" (no code link in the archive), "not matched" (the row's paper URL matched no paper on this site). 33 rows have a graph line, from 14 distinct papers; 25 rows (10 papers) have at least one sample that ran. Counting each paper once: Syntology ran 70 of 211 samples; 141 unverified. Separately, 15 of those 211 are pointer-only (licence): the site points at that code rather than redistributing it, a licence property recorded for ran and unverified samples alike; each cell's tooltip carries the row's own pointer-only count. Read from the graph 2026-09-24. Per-sample status is on the paper page.
Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections