Browse › Computer Vision › Visual Reasoning › Winoground
Winoground Benchmark (Visual Reasoning)
Ability to understand actions and reasoning associated with any visual images
The archive carries no text for this table; the description above is the archive's text for the task Visual Reasoning. 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: Text Score (higher is better), Image Score (higher is better), Group Score (higher is better). Points are placed at the row's paper date; 111 of 114 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 | GPT-4o + CA | 75.5 | 58.5 | 52 | – | Paper | – | 2025 | no code linked | report |
| 2 | GPT-4V (CoT, pick b/w two options) | 75.25 | 68.75 | 58.75 | – | Paper | – | 2023 | no code linked | report |
| 3 | GPT-4V (pick b/w two options) | 69.25 | 46.25 | 39.25 | – | Paper | – | 2023 | no code linked | report |
| 4 | MMICL + CoCoT | 64.25 | 52.5 | 50.75 | – | Paper | Code | 2024 | linked, not harvested | report |
| 5 | GPT-4V + CoCoT | 58.5 | 49.5 | 44.5 | – | Paper | Code | 2024 | linked, not harvested | report |
| 6 | OpenFlamingo + CoCoT | 58.25 | 55.25 | 41.5 | – | Paper | Code | 2024 | linked, not harvested | report |
| 7 | GPT-4V | 54.5 | 42.5 | 37.75 | – | Paper | Code | 2024 | linked, not harvested | report |
| 8 | FIBER (EqSim) | 51.5 | 32.00 | 27.5 | – | Paper | Code | 2023 | 1 of 1 ran · 0 unverified | report |
| 9 | FIBER (finetuned, Flickr30k) | 51.25 | 26.50 | 23.00 | – | Paper | Code | 2023 | 1 of 1 ran · 0 unverified | report |
| 10 | MMICL + CCoT | 51 | 48 | 47.5 | – | Paper | Code | 2024 | linked, not harvested | report |
| 11 | OpenFlamingo + DDCoT | 47.5 | 47.25 | 39 | – | Paper | Code | 2024 | linked, not harvested | report |
| 12 | VQ2 | 47 | 42.2 | 30.5 | – | Paper | Code | 2023 | linked, not harvested | report |
| 13 | MMICL + DDCoT | 46.75 | 45 | 36.75 | – | Paper | Code | 2024 | linked, not harvested | report |
| 14 | X-VLM 16M | 46.7 | 24.5 | 21.2 | – | Paper | Code | 2023 | 2 of 4 ran · 2 unverified | report |
| 15 | PaLI (ft SNLI-VE + Synthetic Data) | 46.5 | 38 | 28.75 | – | Paper | Code | 2023 | linked, not harvested | report |
| 16 | FIBER | 46.25 | 25.75 | 22.25 | – | Paper | Code | 2023 | 1 of 1 ran · 0 unverified | report |
| 17 | MMICL (FLAN-T5-XXL) | 45.50 | 44.99 | 43.00 | – | Paper | Code | 2023 | 5 of 7 ran · 2 unverified | report |
| 18 | PaLI (ft SNLI-VE) | 45.00 | 41.50 | 28.70 | – | Paper | Code | 2023 | linked, not harvested | report |
| 19 | Gemini + DDCoT | 45 | 25 | 23.75 | – | Paper | Code | 2024 | linked, not harvested | report |
| 20 | METER (EqSim) | 45.0 | 22.75 | 18.75 | – | Paper | Code | 2023 | 1 of 1 ran · 0 unverified | report |
| 21 | X-VLM 4M | 44.0 | 26.7 | 21.5 | – | Paper | Code | 2023 | 2 of 4 ran · 2 unverified | report |
| 22 | BLIP2 (ft COCO) | 44.00 | 26.00 | 23.50 | – | Paper | Code | 2023 | linked, not harvested | report |
| 23 | KeyComp* (GPT-4) | 43.5 | 28.7 | 18.2 | – | Paper | Code | 2024 | 1 of 2 ran · 1 unverified | report |
| 24 | METER (finetuned, Flickr30k) | 43.5 | 20.75 | 14.75 | – | Paper | Code | 2023 | 1 of 1 ran · 0 unverified | report |
| 25 | BLIP2 (SGVL) | 42.8 | 28.5 | 23.3 | – | Paper | – | 2023 | no code linked | report |
| 26 | BLIP (SGVL) | 42.8 | 27.3 | 21.5 | – | Paper | – | 2023 | no code linked | report |
| 27 | KeyComp* (GPT-3.5) | 42.7 | 27.8 | 17.4 | – | Paper | Code | 2024 | 1 of 2 ran · 1 unverified | report |
| 28 | OpenFlamingo + CCoT | 42.5 | 27.5 | 20 | – | Paper | Code | 2024 | linked, not harvested | report |
| 29 | NegBLIP | 42.5 | 24.0 | 18.5 | – | Paper | – | 2023 | no code linked | report |
| 30 | IAIS large (Flickr30k) | 42.50 | 19.75 | 16.00 | – | Paper | – | 2022 | no code linked | report |
| 31 | LLaVA-1.5-CCoT | 42.0 | 35.5 | 22.3 | – | Paper | Code | 2023 | 3 of 4 ran · 1 unverified | report |
| 32 | BLIP2 | 42.0 | 23.8 | 19.0 | – | Paper | – | 2023 | no code linked | report |
| 33 | IAIS large (COCO) | 41.75 | 19.75 | 15.50 | – | Paper | – | 2022 | no code linked | report |
| 34 | NegBLIP2 | 41.5 | 26.0 | 20.5 | – | Paper | – | 2023 | no code linked | report |
| 35 | BLIP (+Graph Text, +Graph Neg) | 40.5 | 25.5 | 19.0 | – | Paper | – | 2023 | no code linked | report |
| 36 | BLIP (+Graph Text) | 40.3 | 20.5 | 16.5 | – | Paper | – | 2023 | no code linked | report |
| 37 | Gemini + CoCoT | 40 | 32.5 | 27.75 | – | Paper | Code | 2024 | linked, not harvested | report |
| 38 | CACR base | 39.25 | 17.75 | 14.25 | – | Paper | – | 2022 | no code linked | report |
| 39 | METER | 39.25 | 15.75 | 12.00 | – | Paper | Code | 2023 | 1 of 1 ran · 0 unverified | report |
| 40 | OpenFlamingo | 39 | 41.25 | 33.25 | – | Paper | Code | 2024 | linked, not harvested | report |
| 41 | BLIP | 39.0 | 19.2 | 15.0 | – | Paper | – | 2023 | no code linked | report |
| 42 | GPT-4V (image-caption match answer yes/no, zero-shot) | 38.00 | 38.00 | 38.00 | – | – | – | not matched | report | |
| 43 | UNITER large | 38.00 | 14.00 | 10.50 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 44 | VinVL | 37.75 | 17.75 | 14.50 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 45 | ViLLA large | 37.00 | 13.25 | 11.00 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 46 | BLIP (VisualGPTScore, α-tuned) | 36.5 | 21.5 | 16.8 | – | Paper | Code | 2023 | linked, not harvested | report |
| 47 | BLIP 14M | 36.5 | 18.5 | 14.5 | – | Paper | Code | 2023 | 2 of 4 ran · 2 unverified | report |
| 48 | ViT-B/16 + BERT base + ViLEM | 36.5 | – | Paper | – | 2023 | no code linked | report | ||
| 49 | LLaVA-1.5 | 36.0 | 33.3 | 20.1 | – | Paper | Code | 2023 | 3 of 4 ran · 1 unverified | report |
| 50 | BLIP (ITM) | 35.8 | 15.8 | 13.3 | – | Paper | Code | 2023 | linked, not harvested | report |
| 51 | BLIP 129M | 35.5 | 15.0 | 11.7 | – | Paper | Code | 2023 | 2 of 4 ran · 2 unverified | report |
| 52 | ROSITA (Flickr30k) | 35.25 | 15.25 | 12.25 | – | Paper | – | 2022 | no code linked | report |
| 53 | ViLT (ViT-B/32) | 34.75 | 14.00 | 9.25 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 54 | BLIP 129M (CapFilt/L) | 34.7 | 15.2 | 12.2 | – | Paper | Code | 2023 | 2 of 4 ran · 2 unverified | report |
| 55 | BLIP-ViT/L 129M | 34.7 | 14.5 | 12.2 | – | Paper | Code | 2023 | 2 of 4 ran · 2 unverified | report |
| 56 | Diffusion Classifier (zero-shot) | 34.00 | – | Paper | Code | 2023 | 2 of 2 ran · 0 unverified | report | ||
| 57 | PEVL 14M | 33.2 | 15.7 | 12.2 | – | Paper | Code | 2023 | 2 of 4 ran · 2 unverified | report |
| 58 | ALBEF 14M | 32.5 | 16.2 | 12.7 | – | Paper | Code | 2023 | 2 of 4 ran · 2 unverified | report |
| 59 | FLAVA (ITM) | 32.25 | 20.50 | 14.25 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 60 | UNITER base | 32.25 | 13.25 | 10.00 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 61 | CLIP (SGVL) | 32.0 | 14.0 | 9.8 | – | Paper | – | 2023 | no code linked | report |
| 62 | ViT-B/16 + BERT base | 31.2 | – | Paper | – | 2023 | no code linked | report | ||
| 63 | Gemini | 30.75 | 26 | 25 | – | Paper | Code | 2024 | linked, not harvested | report |
| 64 | OCLIP (ViT-H/14) | 30.75 | 12.75 | – | Paper | – | 2023 | no code linked | report | |
| 65 | CLIP (ViT-B/32) | 30.75 | 10.50 | 8.00 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 66 | OFA large (ITM) | 30.75 | 10.25 | 7.25 | – | Paper | – | 2023 | no code linked | report |
| 67 | KeyComp (GPT-3.5) | 30.3 | 24.6 | 12.4 | – | Paper | Code | 2024 | 1 of 2 ran · 1 unverified | report |
| 68 | CLIP (ViT-L/14) | 30.25 | 8.0 | – | Paper | – | 2023 | no code linked | report | |
| 69 | ViLLA base | 30.00 | 12.00 | 8.00 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 70 | syn-CLIP | 30.00 | 11.50 | 9.50 | – | Paper | Code | 2023 | linked, not harvested | report |
| 71 | syn-CyCLIP | 30.00 | 10.75 | 8.25 | – | Paper | Code | 2023 | linked, not harvested | report |
| 72 | NegCLIP | 29.5 | 10.5 | 8.0 | – | Paper | – | 2023 | no code linked | report |
| 73 | OFA large (TLC-A) | 29.25 | 27.00 | 17.50 | – | Paper | – | 2023 | no code linked | report |
| 74 | ALBEF 4M | 29.2 | 15.5 | 11.0 | – | Paper | Code | 2023 | 2 of 4 ran · 2 unverified | report |
| 75 | LDM-T5 (SelfEval) | 29.00 | 13.50 | – | Paper | – | 2023 | no code linked | report | |
| 76 | CyCLIP | 28.50 | 9.50 | 7.25 | – | Paper | Code | 2023 | linked, not harvested | report |
| 77 | PDM-T5 (SelfEval) | 28.25 | 12.00 | – | Paper | – | 2023 | no code linked | report | |
| 78 | COCA ViT-L14 (f.t on COCO) | 28.25 | 11.50 | 8.25 | – | Paper | Code | 2023 | linked, not harvested | report |
| 79 | LLaVA-1.5-ZS-CoT | 28.0 | 22.5 | 12.3 | – | Paper | Code | 2023 | 3 of 4 ran · 1 unverified | report |
| 80 | BLIP (ITC) | 28.0 | 9.0 | 6.5 | – | Paper | Code | 2023 | linked, not harvested | report |
| 81 | OFA large (ft SNLI-VE) | 27.70 | 14.30 | 9.00 | – | Paper | Code | 2023 | linked, not harvested | report |
| 82 | OFA base (ITM) | 26.75 | 10.75 | 6.50 | – | Paper | – | 2023 | no code linked | report |
| 83 | CLIP RN50x64 | 26.50 | 13.75 | 10.25 | – | Paper | Code | 2023 | linked, not harvested | report |
| 84 | LLaVA-7B (GPTScore) | 25.50 | 17.00 | 10.50 | – | Paper | Code | 2023 | linked, not harvested | report |
| 85 | FLAVA (contrastive) | 25.25 | 13.50 | 9.00 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 86 | Random chance | 25.00 | 25.00 | 16.67 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 87 | LLaVA | 24.8 | 25.0 | 13.0 | – | Paper | – | 2023 | no code linked | report |
| 88 | OFA base (TLC-A) | 24.50 | 23.50 | 13.75 | – | Paper | – | 2023 | no code linked | report |
| 89 | MiniGPT-4-7B (GPTScore) | 24.50 | 21.75 | 11.50 | – | Paper | Code | 2023 | linked, not harvested | report |
| 90 | ViLBERT base | 23.75 | 7.25 | 4.75 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 91 | MiniGPT-4 | 23.3 | 18.0 | 9.5 | – | Paper | – | 2023 | no code linked | report |
| 92 | MiniGPT-4-7B (VisualGPTScore) | 23.25 | 18.00 | 9.50 | – | Paper | Code | 2023 | linked, not harvested | report |
| 93 | VSE++ (COCO, ResNet) | 22.75 | 8.00 | 4.00 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 94 | OFA tiny (ITM) | 22.75 | 7.75 | 4.50 | – | Paper | – | 2023 | no code linked | report |
| 95 | LDM-CLIP (SelfEval) | 22.75 | 7.25 | – | Paper | – | 2023 | no code linked | report | |
| 96 | Gemini + CCoT | 22.5 | 33 | 20.75 | – | Paper | Code | 2024 | linked, not harvested | report |
| 97 | InstructBLIP-CCoT | 21.0 | 21.3 | 8.3 | – | Paper | Code | 2023 | 3 of 4 ran · 1 unverified | report |
| 98 | VSRN (Flickr30k) | 20.00 | 5.00 | 3.50 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 99 | VSE++ (Flickr30k, ResNet) | 20.00 | 5.00 | 2.75 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 100 | VSE++ (Flickr30k, VGG) | 19.75 | 6.25 | 4.50 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 101 | UniT (ITM finetuned) | 19.50 | 6.25 | 4.00 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 102 | LXMERT | 19.25 | 7.00 | 4.00 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 103 | TIFA | 19.00 | 12.50 | 11.30 | – | Paper | Code | 2023 | linked, not harvested | report |
| 104 | IDEFICS 80B | 18.75 | 22.5 | 8.0 | – | – | – | not matched | report | |
| 105 | VSE++ (COCO, VGG) | 18.75 | 5.50 | 3.50 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 106 | VSRN (COCO) | 17.50 | 7.00 | 3.75 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 107 | PDM-CLIP (SelfEval) | 17.00 | 14.00 | – | Paper | – | 2023 | no code linked | report | |
| 108 | IDEFICS 9B | 16.8 | 20.8 | 5.0 | – | – | – | not matched | report | |
| 109 | OFA tiny (TLC-A) | 16.50 | 15.75 | 6.75 | – | Paper | – | 2023 | no code linked | report |
| 110 | VisualBERT base | 15.50 | 2.50 | 1.50 | – | Paper | Code | 2022 | 3 of 4 ran · 1 unverified | report |
| 111 | MiniGPT-4-7B (BERTScore) | 14.00 | 8.00 | 2.75 | – | Paper | Code | 2023 | linked, not harvested | report |
| 112 | LLaVA-7B (BERTScore) | 13.50 | 5.25 | 2.25 | – | Paper | Code | 2023 | linked, not harvested | report |
| 113 | InstructBLIP-ZS-CoT | 9.3 | 16.3 | 4.0 | – | Paper | Code | 2023 | 3 of 4 ran · 1 unverified | report |
| 114 | InstructBLIP | 7.0 | 11.5 | 3.3 | – | Paper | Code | 2023 | 3 of 4 ran · 1 unverified | report |
All 114 rows shown. 111 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). 46 rows have a graph line, from 7 distinct papers; 46 rows (7 papers) have at least one sample that ran. Counting each paper once: Syntology ran 17 of 24 samples; 7 unverified. Separately, 11 of those 24 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