iNaturalist 2021
Version 2021
Models classify nature photographs by species. The 2021 competition uses a new collection of images and species labels.
- Released
- March 2021
- Kind
- General visual questions
- Level
- University
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Subject
- Biological sciences
- Language
- English
- Interaction
- Single turn
- Items from
- Real work and use
- Refresh
- New versions replace it
- Run by
- Anyone
- Based on
- iNaturalist 2018
- In views
- Vision, Biology and medicine
DINOv2 (dst) tops the leaderboard with 87.3%
Best score per entry from SOTA2 Research’s leaderboard, read on November 20th, 2026. 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.
- 87.3 DINOv2 (dst)
- 86.2 Franca
- 82.2 H8(ViT-H/16)
- 81.9 I8(ViT-H/16)
- 81.7 ViT-H/16
- 81.3 H8(ViT-L/16)
- 81.2 I8(ViT-L/16)
- 80.9 ViT-L/16
- 80.7 CAPI
- 77.1 Bootleg
- 76.3 MAE
- 75.1 CrossMAE
- 72.7 data2vec 2.0
- 68.4 DINOv3 ViT-B/16
- 63.0 I-JEPA
- 19.4 dino.txt
- 17.6 DFN
- 17.5 SigLIP
- 10.2 MetaCLIP
- 6.40 EVA-02-CLIP
- 5.70 CLIP
- 1 DINOv2 (dst) Arch: ViT-L/14, Data: LVD, Probe: X-Blk 87.3%
- 2 Franca Arch: ViT-L/14, Data: LAION, Ep.: (3.2), Probe: X-Blk 86.2%
- 3 H8(ViT-H/16) FLOPs: 77.5 x 10^9, Evaluation Protocol: linear 82.2%
- 4 I8(ViT-H/16) FLOPs: 77.7 x 10^9, Evaluation Protocol: linear 81.9%
- 5 ViT-H/16 FLOPs: 127.7 x 10^9, Evaluation Protocol: linear 81.7%
- 6 H8(ViT-L/16) FLOPs: 37.7 x 10^9, Evaluation Protocol: linear 81.3%
- 7 I8(ViT-L/16) FLOPs: 38.5 x 10^9, Evaluation Protocol: linear 81.2%
- 8 ViT-L/16 FLOPs: 61.9 x 10^9, Evaluation Protocol: linear 80.9%
- 9 CAPI Arch: ViT-L/14, Data: IN-1k, Ep.: 6394, Probe: X-Blk 80.7%
- 10 Bootleg Arch: ViT-L/16, Data: IN-1k, Ep.: 600, Probe: X-Blk 77.1%
- 11 MAE Arch: ViT-L/16, Data: IN-1k, Ep.: 1600, Probe: X-Blk 76.3%
- 12 CrossMAE Arch: ViT-L/16, Data: IN-1k, Ep.: 800, Probe: X-Blk 75.1%
- 13 data2vec 2.0 Arch: ViT-L/16, Data: IN-1k, Ep.: 200, Probe: X-Blk 72.7%
- 14 DINOv3 ViT-B/16 Fine-Tuned 68.4%
- 15 I-JEPA Arch: ViT-B/16, Data: IN-1k, Ep.: 600, Probe: X-Blk 63%
- 16 dino.txt Resolution: 336, Pre-training Dataset: LVTD-2.3B, Evaluation Protocol: Zero-shot 19.4%
- 17 DFN Resolution: 224, Pre-training Dataset: DFN-2B, Evaluation Protocol: Zero-shot 17.6%
- 18 SigLIP Resolution: 384, Pre-training Dataset: WebLi, Evaluation Protocol: Zero-shot 17.5%
- 19 MetaCLIP Resolution: 224, Pre-training Dataset: MetaCLIP-2.5B, Evaluation Protocol: Zero-shot 10.2%
- 20 EVA-02-CLIP Resolution: 336, Pre-training Dataset: Merged-2B, Evaluation Protocol: Zero-shot 6.40%
- 21 CLIP Resolution: 336, Pre-training Dataset: WIT-400M, Evaluation Protocol: Zero-shot 5.70%
Source: SOTA2 Research’s leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | DINOv2 (dst) | 87.3% | Arch: ViT-L/14, Data: LVD, Probe: X-Blk | Official leaderboard | |
| 2 | Franca | 86.2% | Arch: ViT-L/14, Data: LAION, Ep.: (3.2), Probe: X-Blk | Official leaderboard | |
| 3 | H8(ViT-H/16) | 82.2% | FLOPs: 77.5 x 10^9, Evaluation Protocol: linear | Official leaderboard | |
| 4 | I8(ViT-H/16) | 81.9% | FLOPs: 77.7 x 10^9, Evaluation Protocol: linear | Official leaderboard | |
| 5 | ViT-H/16 | 81.7% | FLOPs: 127.7 x 10^9, Evaluation Protocol: linear | Official leaderboard | |
| 6 | H8(ViT-L/16) | 81.3% | FLOPs: 37.7 x 10^9, Evaluation Protocol: linear | Official leaderboard | |
| 7 | I8(ViT-L/16) | 81.2% | FLOPs: 38.5 x 10^9, Evaluation Protocol: linear | Official leaderboard | |
| 8 | ViT-L/16 | 80.9% | FLOPs: 61.9 x 10^9, Evaluation Protocol: linear | Official leaderboard | |
| 9 | CAPI | 80.7% | Arch: ViT-L/14, Data: IN-1k, Ep.: 6394, Probe: X-Blk | Official leaderboard | |
| 10 | Bootleg | 77.1% | Arch: ViT-L/16, Data: IN-1k, Ep.: 600, Probe: X-Blk | Official leaderboard | |
| 11 | MAE | 76.3% | Arch: ViT-L/16, Data: IN-1k, Ep.: 1600, Probe: X-Blk | Official leaderboard | |
| 12 | CrossMAE | 75.1% | Arch: ViT-L/16, Data: IN-1k, Ep.: 800, Probe: X-Blk | Official leaderboard | |
| 13 | data2vec 2.0 | 72.7% | Arch: ViT-L/16, Data: IN-1k, Ep.: 200, Probe: X-Blk | Official leaderboard | |
| 14 | DINOv3 ViT-B/16 | 68.4% | Fine-Tuned | Official leaderboard | |
| 15 | I-JEPA | 63% | Arch: ViT-B/16, Data: IN-1k, Ep.: 600, Probe: X-Blk | Official leaderboard | |
| 16 | dino.txt | 19.4% | Resolution: 336, Pre-training Dataset: LVTD-2.3B, Evaluation Protocol: Zero-shot | Official leaderboard | |
| 17 | DFN | 17.6% | Resolution: 224, Pre-training Dataset: DFN-2B, Evaluation Protocol: Zero-shot | Official leaderboard | |
| 18 | SigLIP | 17.5% | Resolution: 384, Pre-training Dataset: WebLi, Evaluation Protocol: Zero-shot | Official leaderboard | |
| 19 | MetaCLIP | 10.2% | Resolution: 224, Pre-training Dataset: MetaCLIP-2.5B, Evaluation Protocol: Zero-shot | Official leaderboard | |
| 20 | EVA-02-CLIP | 6.40% | Resolution: 336, Pre-training Dataset: Merged-2B, Evaluation Protocol: Zero-shot | Official leaderboard | |
| 21 | CLIP | 5.70% | Resolution: 336, Pre-training Dataset: WIT-400M, Evaluation Protocol: Zero-shot | 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.
Versions
More 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 official leaderboard (21 entries).
- Pass Models scored 21 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 SOTA2 Research: 21 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.