Benchmark Almanac

COCO-O

EVA 57.8% tops its leaderboard
Not in model cards

Models detect objects under natural changes in image appearance. The benchmark compares detection on drawings, cartoons, and other visual domains.

Made by
COCO-O authors
Released
July 2023
Kind
Detection and captioning
Level
Everyday
Scored
Checked automatically
Answers
Open-ended
Tasks
All public
Measures
Capability
In and out
Images → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
In views
Vision
53
Entries in its paper, 0 models we track

EVA tops the leaderboard with 57.8%

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

  1. 1 EVA Standard 57.8%
  2. 2 DETA Swin-L 48.5%
  3. 3 GLIP-L Swin-L 48%
  4. 4 GRiT ViT-H 42.9%
  5. 5 DINO Swin-L 42.1%
  6. 6 H-Deformable-DETR Swin-L 41.4%
  7. 7 CBNetV2 Swin-L 39%
  8. 8 SOLQ Swin-L 38.8%
  9. 9 Cascade Mask R-CNN ConvNeXt-XL 37.5%
  10. 10 DyHead Swin-L 35.3%
  11. 11 ViTDet ViT-H 34.3%
  12. 12 ViT-Adapter BEiTv2-L 34.3%
  13. 13 DiffusionDet Swin-B 34.2%
  14. 14 FIBER-B Swin-B 33.7%
  15. 15 QueryInst Swin-L 33.2%
  16. 16 YOLOR-D6 Standard 33%
  17. 17 YOLOv6-L6 Standard 32.5%
  18. 18 YOLOv7-E6E Standard 32%
  19. 19 Det-AdvProp EfficientNet-B5 30.8%
  20. 20 YOLOv4-P6 Standard 30.4%
  21. 21 YOLOX-X Standard 30.3%
  22. 22 CenterNet2 R2-101-DCN 29.5%
  23. 23 YOLOv5l6 Standard 29.3%
  24. 24 GLIP-T Swin-T 29.1%
  25. 25 YOLOv7 Standard 29%
  26. 26 Mask R-CNN MogaNet-L 28.8%
  27. 27 EfficientDet-D5 EfficientNet-B5 28.5%
  28. 28 HTC RX-101-64x4d-DCN 28.4%
  29. 29 VFNet RX-101-64x4d 28%
  30. 30 GCNet RX-101-32x4d-DCN 26%
  31. 31 RetinaNet PVTv2-B2 25.2%
  32. 32 GFLv2 R2-101-DCN 25.1%
  33. 33 Faster R-CNN PVTv2-B2 24.9%
  34. 34 RepPointsV2 RX-101-64x4d-DCN 24.9%
  35. 35 UniverseNet R2-101-DCN 24.8%
  36. 36 TOOD RX-101-64x4d 23.1%
  37. 37 EfficientDet-D1 Det-AdvProp 22.9%
  38. 38 FCOS RX-101-64x4d 21%
  39. 39 YOLOX-S Standard 20.6%
  40. 40 Deformable DETR ResNet-50 20.1%
  41. 41 YOLOS-B ViT-B 20%
  42. 42 VarifocalNet ResNet-50 18.8%
  43. 43 SSD Robustness study baseline 18.7%
  44. 44 Deformable-DETR ResNet-50 18.5%
  45. 45 Cascade R-CNN ResNet-50 18.2%
  46. 46 YOLOv5s Standard 17.8%
  47. 47 Sparse R-CNN ResNet-50 17.5%
  48. 48 NanoDet ShuffleNetV2 17.3%
  49. 49 DETR ResNet-50 17.1%
  50. 50 ATSS ResNet-50 16.8%
  51. 51 YOLACT ResNet-101 16.8%
  52. 52 YOLOS-S Standard 15.5%
  53. 53 YOLOv3 DarkNet-53 14.8%

Source: The COCO-O paper

#EntryScoreSettingDateSource
1 EVA 57.8% Standard Official leaderboard
2 DETA 48.5% Swin-L Official leaderboard
3 GLIP-L 48% Swin-L Official leaderboard
4 GRiT 42.9% ViT-H Official leaderboard
5 DINO 42.1% Swin-L Official leaderboard
6 H-Deformable-DETR 41.4% Swin-L Official leaderboard
7 CBNetV2 39% Swin-L Official leaderboard
8 SOLQ 38.8% Swin-L Official leaderboard
9 Cascade Mask R-CNN 37.5% ConvNeXt-XL Official leaderboard
10 DyHead 35.3% Swin-L Official leaderboard
11 ViTDet 34.3% ViT-H Official leaderboard
12 ViT-Adapter 34.3% BEiTv2-L Official leaderboard
13 DiffusionDet 34.2% Swin-B Official leaderboard
14 FIBER-B 33.7% Swin-B Official leaderboard
15 QueryInst 33.2% Swin-L Official leaderboard
16 YOLOR-D6 33% Standard Official leaderboard
17 YOLOv6-L6 32.5% Standard Official leaderboard
18 YOLOv7-E6E 32% Standard Official leaderboard
19 Det-AdvProp 30.8% EfficientNet-B5 Official leaderboard
20 YOLOv4-P6 30.4% Standard Official leaderboard
21 YOLOX-X 30.3% Standard Official leaderboard
22 CenterNet2 29.5% R2-101-DCN Official leaderboard
23 YOLOv5l6 29.3% Standard Official leaderboard
24 GLIP-T 29.1% Swin-T Official leaderboard
25 YOLOv7 29% Standard Official leaderboard
26 Mask R-CNN 28.8% MogaNet-L Official leaderboard
27 EfficientDet-D5 28.5% EfficientNet-B5 Official leaderboard
28 HTC 28.4% RX-101-64x4d-DCN Official leaderboard
29 VFNet 28% RX-101-64x4d Official leaderboard
30 GCNet 26% RX-101-32x4d-DCN Official leaderboard
31 RetinaNet 25.2% PVTv2-B2 Official leaderboard
32 GFLv2 25.1% R2-101-DCN Official leaderboard
33 Faster R-CNN 24.9% PVTv2-B2 Official leaderboard
34 RepPointsV2 24.9% RX-101-64x4d-DCN Official leaderboard
35 UniverseNet 24.8% R2-101-DCN Official leaderboard
36 TOOD 23.1% RX-101-64x4d Official leaderboard
37 EfficientDet-D1 22.9% Det-AdvProp Official leaderboard
38 FCOS 21% RX-101-64x4d Official leaderboard
39 YOLOX-S 20.6% Standard Official leaderboard
40 Deformable DETR 20.1% ResNet-50 Official leaderboard
41 YOLOS-B 20% ViT-B Official leaderboard
42 VarifocalNet 18.8% ResNet-50 Official leaderboard
43 SSD 18.7% Robustness study baseline Official leaderboard
44 Deformable-DETR 18.5% ResNet-50 Official leaderboard
45 Cascade R-CNN 18.2% ResNet-50 Official leaderboard
46 YOLOv5s 17.8% Standard Official leaderboard
47 Sparse R-CNN 17.5% ResNet-50 Official leaderboard
48 NanoDet 17.3% ShuffleNetV2 Official leaderboard
49 DETR 17.1% ResNet-50 Official leaderboard
50 ATSS 16.8% ResNet-50 Official leaderboard
51 YOLACT 16.8% ResNet-101 Official leaderboard
52 YOLOS-S 15.5% Standard Official leaderboard
53 YOLOv3 14.8% DarkNet-53 Official leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Vision

All 636 in Vision
  • MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy. Qwen 3.6 Plus 86%
  • MMMU-Pro In use Harder MMMU: college-level multimodal questions with more options and vision-only inputs to remove text shortcuts. GPT-5.6 Sol 84.6%
  • CharXiv In use Chart understanding on real arXiv figures. Reasoning questions require synthesizing information from complex charts. Claude Fable 5 93.5%
  • MathVista In use Math reasoning in visual contexts (charts, geometry, puzzles) from 31 datasets. Answer accuracy. Seed2.1 Pro 90.7%
  • MATH-Vision (MathVision) In use 3,040 competition math problems with visual context. Answer accuracy. Qwen 3.8 Max 97.7%
  • MMStar In use Models answer 1,500 human-curated image questions that cannot be solved without the image, across 6 core capabilities. Scored by accuracy. Qwen 3.8 Max 85.9%

Data audit

All 8 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
  • Pass Independent runs Scores from the benchmark’s paper (53 entries).
  • Pass Models scored 53 entries on its leaderboard, none a model we track.
  • Note Newest score No scores yet.
  • Note Frontier models No models scored yet.
  • Note Headroom No scores yet.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 53 entries at the last read, none a model we track.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check