GEO-Bench-2
Models interpret Earth imagery across 19 tasks, including field boundaries, forest biomass and tree species. The authors normalize task scores to compare model capabilities.
The leaderboard publishes results for 18 distinct model backbones, including DINOv3. The authors aggregate normalized results by capability.
- Made by
- AI Alliance GEO-Bench-2 team
- Released
- November 2025
- Kind
- Spatial reasoning and counting
- Level
- University
- Scored
- Checked automatically
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Capability
- Language
- English
- In and out
- Maps and geodata, Images → Text, Images, Maps and geodata
- Interaction
- Single turn
- Items from
- Another benchmark
- Refresh
- New versions replace it
- Run by
- Anyone
- Based on
- A version of another benchmark
- In views
- Vision, Image, video and 3D generation
Clay-V1 ViT-B tops the leaderboard with 78.6
Best score per entry from the benchmark's own 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.
- 78.6 Clay-V1 ViT-B
- 77.8 ConvNext-XLarge-ImageNet
- 70.8 DinoV3-ConvNext-Large-WEB
- 69.9 ConvNext-Large-ImageNet
- 68.9 Prithvi-EO-V2 600M TL
- 67.4 TerraMind-V1-Large
- 65.6 Satlas-Swin 100M
- 61.9 DOFA-ViT 300M
- 59.4 Prithvi-EO-V2 300M TL
- 58.1 DinoV3-ViT-L-SAT
- 54.7 TerraMind-V1-Base
- 52.6 Satlas-SwinB-Naip
- 28.9 Resnet50-DeCUR
- 26.8 Resnet50-ImageNet
- 1 Clay-V1 ViT-B Fully fine-tuned, submission 2ce4a907 78.6
- 2 ConvNext-XLarge-ImageNet Fully fine-tuned, submission 114be1f0 77.8
- 3 DinoV3-ConvNext-Large-WEB Fully fine-tuned, submission 114be1f0 70.8
- 4 ConvNext-Large-ImageNet Fully fine-tuned, submission 114be1f0 69.9
- 5 Prithvi-EO-V2 600M TL Fully fine-tuned, submission 114be1f0 68.9
- 6 TerraMind-V1-Large Fully fine-tuned, submission 2ce4a907 67.4
- 7 Satlas-Swin 100M Fully fine-tuned, submission 114be1f0 65.6
- 8 DOFA-ViT 300M Fully fine-tuned, submission 2ce4a907 61.9
- 9 Prithvi-EO-V2 300M TL Fully fine-tuned, submission 114be1f0 59.4
- 10 DinoV3-ViT-L-SAT Fully fine-tuned, submission 114be1f0 58.1
- 11 TerraMind-V1-Base Fully fine-tuned, submission 114be1f0 54.7
- 12 Satlas-SwinB-Naip Fully fine-tuned, submission 114be1f0 52.6
- 13 Resnet50-DeCUR Fully fine-tuned, submission 114be1f0 28.9
- 14 Resnet50-ImageNet Fully fine-tuned, submission 114be1f0 26.8
Source: Official GEO-Bench-2 leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Clay-V1 ViT-B 0.1B | 78.6 | Fully fine-tuned, submission 2ce4a907 | Official leaderboard | |
| 2 | ConvNext-XLarge-ImageNet 0.4B | 77.8 | Fully fine-tuned, submission 114be1f0 | Official leaderboard | |
| 3 | DinoV3-ConvNext-Large-WEB 0.2B | 70.8 | Fully fine-tuned, submission 114be1f0 | Official leaderboard | |
| 4 | ConvNext-Large-ImageNet 0.2B | 69.9 | Fully fine-tuned, submission 114be1f0 | Official leaderboard | |
| 5 | Prithvi-EO-V2 600M TL 0.6B | 68.9 | Fully fine-tuned, submission 114be1f0 | Official leaderboard | |
| 6 | TerraMind-V1-Large 0.3B | 67.4 | Fully fine-tuned, submission 2ce4a907 | Official leaderboard | |
| 7 | Satlas-Swin 100M 0.1B | 65.6 | Fully fine-tuned, submission 114be1f0 | Official leaderboard | |
| 8 | DOFA-ViT 300M 0.3B | 61.9 | Fully fine-tuned, submission 2ce4a907 | Official leaderboard | |
| 9 | Prithvi-EO-V2 300M TL 0.3B | 59.4 | Fully fine-tuned, submission 114be1f0 | Official leaderboard | |
| 10 | DinoV3-ViT-L-SAT 0.3B | 58.1 | Fully fine-tuned, submission 114be1f0 | Official leaderboard | |
| 11 | TerraMind-V1-Base 0.1B | 54.7 | Fully fine-tuned, submission 114be1f0 | Official leaderboard | |
| 12 | Satlas-SwinB-Naip 0.1B | 52.6 | Fully fine-tuned, submission 114be1f0 | Official leaderboard | |
| 13 | Resnet50-DeCUR 0B | 28.9 | Fully fine-tuned, submission 114be1f0 | Official leaderboard | |
| 14 | Resnet50-ImageNet 0B | 26.8 | Fully fine-tuned, submission 114be1f0 | 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
- 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 (14 entries).
- Pass Models scored 14 entries on its leaderboard, none a model we track.
- Note Newest score No scores yet.
- Note Frontier models No models scored yet.
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on huggingface.co: 14 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.