NautData test set
Models answer questions, describe underwater scenes, and locate organisms in images. The paper reports 7,916 test examples on 3,920 images.
Passes test 4. The USI-Brain team evaluated the public split. Both released test files contain 13,916 examples. The papers label the visual question score differently.
- Released
- October 2025
- Kind
- General visual questions
- Level
- University
- Scored
- Checked automatically
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Capability
- Subject
- Biological sciences
- Language
- English
- Interaction
- Single turn
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Vision, Biology and medicine
- Strongest model tested
-
Gemini 2.0 Flash Feb 2025 #147 on the capability index
- Last model added
- February 2025
- Scores refreshed
- November 20th, 2026
mPLUG-Owl3-7B leads with 91.9% among models released by February 2025 mPLUG-Owl3-7B leads with 91.9% among models released by February 2025
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper reports distinct task metrics. The headline is coarse classification accuracy, its first column. 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. This view shows only models we track, each at its best.
The newest model with a score here came out in February 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1
mPLUG-Owl3-7B Instruction-tuning 91.9%
- 2
InternVL2.5-8B Instruction-tuning 91.3%
- 3 Nautilus (LLaVA-1.5) Instruction-tuning 91%
- 4 Nautilus (Qwen2.5-VL) Instruction-tuning 90.3%
- 5 llava-v1.5-7b Instruction-tuning 90%
- 6
Qwen2.5-VL-7B Instruction-tuning 85.3%
- 7 MiniGPTv2 Instruction-tuning 80%
- 8
Gemini 2.0 Flash Feb 2025, Zero-shot 55.5%
- 9 GPT-4o (snapshot unspecified) Zero-shot 55.2%
- 10
Qwen2.5-VL-72B Zero-shot 55.2%
Source: The NautData test set paper
- 1
mPLUG-Owl3-7B Instruction-tuning 91.9%
- 2
InternVL2.5-8B Instruction-tuning 91.3%
- 3 llava-v1.5-7b Instruction-tuning 90%
- 4
Qwen2.5-VL-7B Instruction-tuning 85.3%
- 5
Gemini 2.0 Flash Feb 2025, Zero-shot 55.5%
- 6
Qwen2.5-VL-72B Zero-shot 55.2%
Source: The NautData test set paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 91.9% | Instruction-tuning | Aug 2024 | Official leaderboard |
| 2 | | 91.3% | Instruction-tuning | Dec 2024 | Official leaderboard |
| 3 | Nautilus (LLaVA-1.5) | 91% | Instruction-tuning | Official leaderboard | |
| 4 | Nautilus (Qwen2.5-VL) | 90.3% | Instruction-tuning | Official leaderboard | |
| 5 | llava-v1.5-7b | 90% | Instruction-tuning | Oct 2023 | Official leaderboard |
| 6 | | 85.3% | Instruction-tuning | Jan 2025 | Official leaderboard |
| 7 | MiniGPTv2 | 80% | Instruction-tuning | Official leaderboard | |
| 8 | | 55.5% | Zero-shot | Feb 2025 | Official leaderboard |
| 9 | GPT-4o (snapshot unspecified) | 55.2% | Zero-shot | Official leaderboard | |
| 10 | | 55.2% | Zero-shot | Jan 2025 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 91.9% | Instruction-tuning | Aug 2024 | Official leaderboard |
| 2 | | 91.3% | Instruction-tuning | Dec 2024 | Official leaderboard |
| 3 | llava-v1.5-7b | 90% | Instruction-tuning | Oct 2023 | Official leaderboard |
| 4 | | 85.3% | Instruction-tuning | Jan 2025 | Official leaderboard |
| 5 | | 55.5% | Zero-shot | Feb 2025 | Official leaderboard |
| 6 | | 55.2% | Zero-shot | Jan 2025 | Official leaderboard |
Rankings by part
We read 9 parts from the benchmark’s paper, each ranking models on one part of the test, such as Coarse accuracy, Fine accuracy and Image caption METEOR. Qwen2.5-VL-7B leads 5 of them, and mPLUG-Owl3-7B leads the other 4. Arrows show how far a model moved from its overall rank among the models we track.
Coarse accuracy
6 models we track have a score here.
- 1
mPLUG-Owl3-7B 91.9%
- 2
InternVL2.5-8B 91.3%
- 3 llava-v1.5-7b 90%
- 4
Qwen2.5-VL-7B 85.3%
- 5
Gemini 2.0 Flash Feb 2025 55.5%
- 6
Qwen2.5-VL-72B 55.2%
Fine accuracy
6 models we track have a score here.
- 1
mPLUG-Owl3-7B 92%
- 2
InternVL2.5-8B 90.4%
- 3 llava-v1.5-7b 89.8%
- 4
Qwen2.5-VL-7B 88.2%
- 5
Gemini 2.0 Flash Feb 2025 54.3%
- 6
Qwen2.5-VL-72B 54.2%
Image caption METEOR
6 models we track have a score here.
- 1
Qwen2.5-VL-7B ↑3 0.22
- 2
mPLUG-Owl3-7B ↓1 0.22
- 3 llava-v1.5-7b 0.21
- 4
InternVL2.5-8B ↓2 0.21
- 5
Gemini 2.0 Flash Feb 2025 0.18
- 6
Qwen2.5-VL-72B 0.17
Region caption METEOR
6 models we track have a score here.
- 1
mPLUG-Owl3-7B 0.21
- 2
Qwen2.5-VL-7B ↑2 0.20
- 3
InternVL2.5-8B ↓1 0.20
- 4 llava-v1.5-7b ↓1 0.19
- 5
Gemini 2.0 Flash Feb 2025 0.14
- 6
Qwen2.5-VL-72B 0.13
Grounding mIoU
6 models we track have a score here.
- 1
Qwen2.5-VL-7B ↑3 52.5%
- 2
InternVL2.5-8B 50.4%
- 3 llava-v1.5-7b 43.5%
- 4
Qwen2.5-VL-72B ↑2 42.3%
- 5
mPLUG-Owl3-7B ↓4 41.1%
- 6
Gemini 2.0 Flash Feb 2025 ↓1 21.2%
Grounding PR@0.5
6 models we track have a score here.
- 1
Qwen2.5-VL-7B ↑3 57.6%
- 2
InternVL2.5-8B 54.6%
- 3 llava-v1.5-7b 48.2%
- 4
Qwen2.5-VL-72B ↑2 46.4%
- 5
mPLUG-Owl3-7B ↓4 45.7%
- 6
Gemini 2.0 Flash Feb 2025 ↓1 20.6%
Detection mAP
6 models we track have a score here.
- 1
Qwen2.5-VL-7B ↑3 24.5%
- 2
InternVL2.5-8B 18.3%
- 3
mPLUG-Owl3-7B ↓2 10.3%
- 4 llava-v1.5-7b ↓1 9.80%
- 5
Qwen2.5-VL-72B ↑1 8.80%
- 6
Gemini 2.0 Flash Feb 2025 ↓1 2.40%
Detection mAP@0.5
6 models we track have a score here.
- 1
Qwen2.5-VL-7B ↑3 41.7%
- 2
InternVL2.5-8B 30.5%
- 3
mPLUG-Owl3-7B ↓2 23.3%
- 4 llava-v1.5-7b ↓1 19%
- 5
Qwen2.5-VL-72B ↑1 14.7%
- 6
Gemini 2.0 Flash Feb 2025 ↓1 7.80%
VQA METEOR
6 models we track have a score here.
- 1
mPLUG-Owl3-7B 0.38
- 2
InternVL2.5-8B 0.38
- 3
Qwen2.5-VL-7B ↑1 0.38
- 4 llava-v1.5-7b ↓1 0.36
- 5
Gemini 2.0 Flash Feb 2025 0.22
- 6
Qwen2.5-VL-72B 0.22
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The NautData test set paper
Llava-v1.5-7b on the official leaderboard passed 85% in October 2023
Models released before it came out already scored that high. All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official NautData test set 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 benchmark’s paper (10 entries).
- Pass Models scored 6 models we track, plus 4 other entries on its leaderboard.
- Note Newest score The newest came 21 months ago, in February 2025, for Gemini 2.0 Flash. That’s expected for a saturated test.
- Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is Gemini 2.0 Flash (#147).
- Note Headroom Saturated. The best is 91.9%, past 85% since October 2023.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 6 rows for models we track and 4 other entries at the last read.
- 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.