Recent studyBlind TTS Elo is live. Compare two anonymous voice samples, vote after listening, and help separate real preference signal from noise.Vote in the study ->
Codesota · Benchmark · CIFAR-100Home/Leaderboards/Vision & Documents/Image Classification/CIFAR-100
Unknown

CIFAR-100.

60K 32x32 color images in 100 fine-grained classes grouped into 20 superclasses. More challenging than CIFAR-10.

Paper ↗Leaderboard ↓
§ 01 · SOTA history

Year over year.

§ 02 · Leaderboard

Results by metric.

Found a wrong score or missing run?
Use row edits to send a sourced correction into moderation.
Add / edit result ↗Report issue ↗

accuracy

Accuracy is the reported evaluation metric for CIFAR-100. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for accuracyverifiedpapervendorcommunityunverified
RankModelTrustScoreYearLinksEdit
01EVA-02-L
EVA-02-L/14+ fine-tuned on CIFAR-100. Pre-trained with EVA-CLIP on Objects365 + ImageNet-21K. State-of-the-art as of 2023. arXiv Mar 2023.
paper97.152023Source ↗Edit result
02CoAtNet-7
CoAtNet-7 (2.4B params) fine-tuned on CIFAR-100. Pre-trained on ImageNet-21K. arXiv Jun 2021, NeurIPS 2021.
paper96.382021Source ↗Edit result
03ConvNeXt V2-H
ConvNeXt V2-H fine-tuned on CIFAR-100 after FCMAE pre-training on ImageNet-22K. arXiv Jan 2023, CVPR 2023.
paper96.172023Source ↗Edit result
04MAE ViT-H/14
ViT-H/14 fine-tuned on CIFAR-100 after MAE pre-training on ImageNet-1K. arXiv Nov 2021, CVPR 2022.
paper96.082021Source ↗Edit result
05SwinV2-G
SwinV2-G (3B params) fine-tuned on CIFAR-100. Pre-trained on ImageNet-21K with resolution 192^2. arXiv Nov 2021, CVPR 2022.
paper96.012021Source ↗Edit result
06DeiT III-H/14
DeiT III ViT-H/14 fine-tuned on CIFAR-100. Improved training recipe for ViTs. arXiv Apr 2022, ECCV 2022.
paper95.942022Source ↗Edit result
07InternImage-XL
InternImage-XL fine-tuned on CIFAR-100. Uses deformable convolutions as core operator. arXiv Nov 2022, CVPR 2023.
paper95.772022Source ↗Edit result
08FasterViT-6
FasterViT-6 fine-tuned on CIFAR-100. Hierarchical ViT with carrier tokens for high-resolution efficiency. arXiv Jun 2023, ICLR 2024.
paper95.722023Source ↗Edit result
09Vision Transformer (ViT-H/14)unverified94.552020Paper ↗Code ↗Edit result
10vit-h-14
Fine-tuned from ImageNet pretraining.
paper94.552025Source ↗Edit result
11ViT-H/14
Fine-tuned from ImageNet pretraining.
unverified94.552025Source ↗Edit result
12AIMv2 ViT-3B/14 448pxunverified94.52024Paper ↗Code ↗Edit result
13AIMv2-3B
AIMv2-3B (2.7B params), multimodal autoregressive pre-training, patch14 448px. 94.5% on CIFAR-100 using attentive probing (frozen backbone). Apple, presented at CVPR 2025. Source: official HF model card. Paper: arxiv:2411.14402, Nov 2024.
verified94.52026Source ↗Edit result
14AIMv2-1B
AIMv2-1B, multimodal autoregressive pre-training, patch14 224px. 94.1% on CIFAR-100 using attentive probing (frozen backbone). Apple, presented at CVPR 2025. Source: official HF model card. Paper: arxiv:2411.14402, Nov 2024.
verified94.12026Source ↗Edit result
15BiT-Lunverified93.512019Paper ↗Code ↗Edit result
16ViT-L/16 (IN-21K)
Vision Transformer ViT-L/16, pretrained on ImageNet-21K and finetuned on CIFAR-100. 93.25% reported in ViT paper (Table 5). Paper: Dosovitskiy et al. 2021, arxiv:2010.11929.
verified93.252026Source ↗Edit result
17BEiTunverified91.82021Paper ↗Code ↗Edit result
18efficientnet-b7
Transfer learning from ImageNet.
paper91.72025Source ↗Edit result
19ViT-B/16
Fine-tuned from ImageNet-21K.
unverified91.482025Source ↗Edit result
20vit-b-16
Fine-tuned from ImageNet-21K.
paper91.482025Source ↗Edit result
21LeJEPA ViT-L (304M)unverified83.712025Paper ↗Code ↗Edit result
22CN-CLIPunverified79.72022Paper ↗Code ↗Edit result
23ResNet-50
With Cutout augmentation.
paper78.042025Source ↗Edit result
§ 04 · Submit a result

Add to the leaderboard.

← Back to Image Classification