DatBench
Models answer visual questions drawn from 33 datasets. The suite reports separate scores across nine visual capabilities.
The January 2026 paper reports 27 models by capability. GLM-4.1V-9B-Thinking led the chart subset at 76.0%.
- Made by
- DatologyAI
- Released
- January 2026
- Kind
- General visual questions
- Level
- University
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Vision
- Last model added
- October 2025
- Scores refreshed
- November 20th, 2026
Qwen3-VL-4B-Instruct leads with 64% among models released by October 2025 Qwen3-VL-4B-Instruct leads with 64% among models released by October 2025
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. This is sample-weighted accuracy calculated from the rounded DatBench capability scores and released test-set sizes. 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 October 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
- 64.0
Qwen3-VL-4B-Instruct
- 58.7 GLM-4.1V-9B-Base
- 58.3 GLM-4.1V-9B-Thinking
- 58.1
Qwen3-VL-8B-Instruct
- 54.7 R-4B
- 53.2
Qwen2.5-VL-7B
- 52.3
Qwen3-VL-2B-Instruct
- 51.6
Qwen3-VL-8B-Thinking
- 51.2
InternVL3-9B
- 49.7
InternVL3.5-8B
- 49.6
Qwen2.5-VL-3B
- 48.7
Qwen3-VL-4B-Thinking
- 47.1
InternVL3.5-4B
- 47.0
Qwen2.5-Omni-7B
- 45.5
InternVL2.5-8B
- 44.8
InternVL2.5-4B
- 43.9
Qwen3-VL-2B-Thinking
- 41.2
InternVL2-8B
- 39.5
InternVL3-2B
- 38.5 Qwen2.5-Omni-3B
- 36.6 InternVL3_5-2B-Instruct
- 34.3
InternVL2-4B
- 32.9
InternVL2.5-2B
- 31.4 Phi-3.5-vision-instruct
- 25.1
InternVL2-2B
- 15.0
Gemma 3 4B
- 1
Qwen3-VL-4B-Instruct 64%
- 2 GLM-4.1V-9B-Base 58.7%
- 3 GLM-4.1V-9B-Thinking 58.3%
- 4
Qwen3-VL-8B-Instruct 58.1%
- 5 R-4B 54.7%
- 6
Qwen2.5-VL-7B 53.2%
- 7
Qwen3-VL-2B-Instruct 52.3%
- 8
Qwen3-VL-8B-Thinking 51.6%
- 9
InternVL3-9B 51.2%
- 10
InternVL3.5-8B 49.7%
- 11
Qwen2.5-VL-3B 49.6%
- 12
Qwen3-VL-4B-Thinking 48.7%
- 13
InternVL3.5-4B 47.1%
- 14
Qwen2.5-Omni-7B 47%
- 15
InternVL2.5-8B 45.5%
- 16
InternVL2.5-4B 44.8%
- 17
Qwen3-VL-2B-Thinking 43.9%
- 18
InternVL2-8B 41.2%
- 19
InternVL3-2B 39.5%
- 20 Qwen2.5-Omni-3B 38.5%
- 21 InternVL3_5-2B-Instruct 36.6%
- 22
InternVL2-4B 34.3%
- 23
InternVL2.5-2B 32.9%
- 24 Phi-3.5-vision-instruct 31.4%
- 25
InternVL2-2B 25.1%
- 26
Gemma 3 4B 15%
Source: The DatBench paper
- 64.0
Qwen3-VL-4B-Instruct
- 58.3 GLM-4.1V-9B-Thinking
- 58.1
Qwen3-VL-8B-Instruct
- 53.2
Qwen2.5-VL-7B
- 52.3
Qwen3-VL-2B-Instruct
- 51.6
Qwen3-VL-8B-Thinking
- 51.2
InternVL3-9B
- 49.7
InternVL3.5-8B
- 49.6
Qwen2.5-VL-3B
- 48.7
Qwen3-VL-4B-Thinking
- 47.1
InternVL3.5-4B
- 47.0
Qwen2.5-Omni-7B
- 45.5
InternVL2.5-8B
- 44.8
InternVL2.5-4B
- 43.9
Qwen3-VL-2B-Thinking
- 41.2
InternVL2-8B
- 39.5
InternVL3-2B
- 34.3
InternVL2-4B
- 32.9
InternVL2.5-2B
- 31.4 Phi-3.5-vision-instruct
- 25.1
InternVL2-2B
- 15.0
Gemma 3 4B
- 1
Qwen3-VL-4B-Instruct 64%
- 2 GLM-4.1V-9B-Thinking 58.3%
- 3
Qwen3-VL-8B-Instruct 58.1%
- 4
Qwen2.5-VL-7B 53.2%
- 5
Qwen3-VL-2B-Instruct 52.3%
- 6
Qwen3-VL-8B-Thinking 51.6%
- 7
InternVL3-9B 51.2%
- 8
InternVL3.5-8B 49.7%
- 9
Qwen2.5-VL-3B 49.6%
- 10
Qwen3-VL-4B-Thinking 48.7%
- 11
InternVL3.5-4B 47.1%
- 12
Qwen2.5-Omni-7B 47%
- 13
InternVL2.5-8B 45.5%
- 14
InternVL2.5-4B 44.8%
- 15
Qwen3-VL-2B-Thinking 43.9%
- 16
InternVL2-8B 41.2%
- 17
InternVL3-2B 39.5%
- 18
InternVL2-4B 34.3%
- 19
InternVL2.5-2B 32.9%
- 20 Phi-3.5-vision-instruct 31.4%
- 21
InternVL2-2B 25.1%
- 22
Gemma 3 4B 15%
Source: The DatBench paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 64% | Oct 2025 | Official leaderboard | |
| 2 | GLM-4.1V-9B-Base | 58.7% | Official leaderboard | ||
| 3 | GLM-4.1V-9B-Thinking | 58.3% | Jul 2025 | Official leaderboard | |
| 4 | | 58.1% | Oct 2025 | Official leaderboard | |
| 5 | R-4B | 54.7% | Official leaderboard | ||
| 6 | | 53.2% | Jan 2025 | Official leaderboard | |
| 7 | | 52.3% | Oct 2025 | Official leaderboard | |
| 8 | | 51.6% | Oct 2025 | Official leaderboard | |
| 9 | | 51.2% | Apr 2025 | Official leaderboard | |
| 10 | | 49.7% | Aug 2025 | Official leaderboard | |
| 11 | | 49.6% | Jan 2025 | Official leaderboard | |
| 12 | | 48.7% | Oct 2025 | Official leaderboard | |
| 13 | | 47.1% | Aug 2025 | Official leaderboard | |
| 14 | | 47% | Mar 2025 | Official leaderboard | |
| 15 | | 45.5% | Dec 2024 | Official leaderboard | |
| 16 | | 44.8% | Dec 2024 | Official leaderboard | |
| 17 | | 43.9% | Oct 2025 | Official leaderboard | |
| 18 | | 41.2% | Jul 2024 | Official leaderboard | |
| 19 | | 39.5% | Apr 2025 | Official leaderboard | |
| 20 | Qwen2.5-Omni-3B | 38.5% | Official leaderboard | ||
| 21 | InternVL3_5-2B-Instruct | 36.6% | Official leaderboard | ||
| 22 | | 34.3% | Jul 2024 | Official leaderboard | |
| 23 | | 32.9% | Dec 2024 | Official leaderboard | |
| 24 | Phi-3.5-vision-instruct | 31.4% | Aug 2024 | Official leaderboard | |
| 25 | | 25.1% | Jul 2024 | Official leaderboard | |
| 26 | | 15% | Mar 2025 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 64% | Oct 2025 | Official leaderboard | |
| 2 | GLM-4.1V-9B-Thinking | 58.3% | Jul 2025 | Official leaderboard | |
| 3 | | 58.1% | Oct 2025 | Official leaderboard | |
| 4 | | 53.2% | Jan 2025 | Official leaderboard | |
| 5 | | 52.3% | Oct 2025 | Official leaderboard | |
| 6 | | 51.6% | Oct 2025 | Official leaderboard | |
| 7 | | 51.2% | Apr 2025 | Official leaderboard | |
| 8 | | 49.7% | Aug 2025 | Official leaderboard | |
| 9 | | 49.6% | Jan 2025 | Official leaderboard | |
| 10 | | 48.7% | Oct 2025 | Official leaderboard | |
| 11 | | 47.1% | Aug 2025 | Official leaderboard | |
| 12 | | 47% | Mar 2025 | Official leaderboard | |
| 13 | | 45.5% | Dec 2024 | Official leaderboard | |
| 14 | | 44.8% | Dec 2024 | Official leaderboard | |
| 15 | | 43.9% | Oct 2025 | Official leaderboard | |
| 16 | | 41.2% | Jul 2024 | Official leaderboard | |
| 17 | | 39.5% | Apr 2025 | Official leaderboard | |
| 18 | | 34.3% | Jul 2024 | Official leaderboard | |
| 19 | | 32.9% | Dec 2024 | Official leaderboard | |
| 20 | Phi-3.5-vision-instruct | 31.4% | Aug 2024 | Official leaderboard | |
| 21 | | 25.1% | Jul 2024 | Official leaderboard | |
| 22 | | 15% | Mar 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 Chart Understanding, Document Understanding and Scene OCR. GLM-4.1V-9B-Thinking leads 2 of them, and 5 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Chart Understanding
22 models we track have a score here.
- 1 GLM-4.1V-9B-Thinking ↑1 76%
- 2
Qwen3-VL-8B-Thinking ↑4 73.2%
- 3
Qwen3-VL-8B-Instruct 70.5%
- 4
Qwen3-VL-4B-Thinking ↑6 68.9%
- 5
Qwen2.5-VL-7B ↓1 64.1%
- 6
Qwen3-VL-4B-Instruct ↓5 63.8%
- 7
Qwen3-VL-2B-Thinking ↑8 62.3%
- 8
InternVL3.5-8B 60.4%
- 9
InternVL3.5-4B ↑2 58.8%
- 10
InternVL3-9B ↓3 54.8%
Document Understanding
22 models we track have a score here.
- 1
Qwen2.5-VL-3B ↑8 71.1%
- 2
Qwen3-VL-4B-Instruct ↓1 71%
- 3
Qwen2.5-Omni-7B ↑9 70.1%
- 4
Qwen2.5-VL-7B 62.8%
- 5
Qwen3-VL-2B-Instruct 62.7%
- 6 Phi-3.5-vision-instruct ↑14 60.7%
- 7
Qwen3-VL-8B-Instruct ↓4 57.2%
- 8
InternVL2.5-8B ↑5 41.4%
- 9
InternVL3.5-8B ↓1 37%
- 10
InternVL3-9B ↓3 32.3%
Scene OCR
22 models we track have a score here.
- 1
Qwen2.5-VL-7B ↑3 82.6%
- 2
Qwen3-VL-4B-Instruct ↓1 77.9%
- 3
Qwen2.5-Omni-7B ↑9 76.8%
- 4
Qwen2.5-VL-3B ↑5 75.1%
- 5
Qwen3-VL-2B-Instruct 72.3%
- 6
InternVL3-9B ↑1 66%
- 7 GLM-4.1V-9B-Thinking ↓5 59%
- 8
InternVL3-2B ↑9 54.5%
- 9
Qwen3-VL-8B-Instruct ↓6 54.4%
- 10
InternVL3.5-8B ↓2 53.3%
Grounding
22 models we track have a score here.
- 1
Qwen3-VL-8B-Thinking ↑5 88.8%
- 2 GLM-4.1V-9B-Thinking 87.9%
- 3
Qwen3-VL-4B-Thinking ↑7 87.3%
- 4
Qwen3-VL-4B-Instruct ↓3 86.2%
- 5
Qwen3-VL-8B-Instruct ↓2 84.9%
- 6
Qwen3-VL-2B-Thinking ↑9 84.6%
- 7
Qwen3-VL-2B-Instruct ↓2 76.2%
- 8
InternVL2-8B ↑8 73.9%
- 9
InternVL2.5-8B ↑4 70.9%
- 10
InternVL3-9B ↓3 70.9%
Counting
22 models we track have a score here.
- 1
Qwen3-VL-8B-Instruct ↑2 92.6%
- 2
Qwen3-VL-8B-Thinking ↑4 92.5%
- 3
Qwen2.5-VL-7B ↑1 92.4%
- 4 GLM-4.1V-9B-Thinking ↓2 92.3%
- 5
Qwen3-VL-4B-Thinking ↑5 92.3%
- 6
Qwen3-VL-4B-Instruct ↓5 92.2%
- 7
InternVL3-9B 91.4%
- 8
Qwen3-VL-2B-Thinking ↑7 91.2%
- 9
InternVL3.5-8B ↓1 91%
- 10
InternVL3.5-4B ↑1 90.6%
Spatial Reasoning
22 models we track have a score here.
- 1
InternVL3-9B ↑6 32.9%
- 2
Qwen2.5-VL-3B ↑7 27.5%
- 3
InternVL2.5-4B ↑11 25.4%
- 4
InternVL3.5-8B ↑4 24.1%
- 5
InternVL3-2B ↑12 22%
- 6
InternVL2.5-8B ↑7 21.8%
- 7
Qwen3-VL-4B-Instruct ↓6 21.8%
- 8
InternVL3.5-4B ↑3 21.4%
- 9
Qwen2.5-Omni-7B ↑3 21.2%
- 10
Qwen2.5-VL-7B ↓6 20.8%
Math & Logic
22 models we track have a score here.
- 1
Qwen3-VL-8B-Thinking ↑5 43.3%
- 2
Qwen3-VL-4B-Thinking ↑8 38.2%
- 3
Qwen3-VL-8B-Instruct 35.9%
- 4
Qwen3-VL-4B-Instruct ↓3 35.2%
- 5 GLM-4.1V-9B-Thinking ↓3 32.4%
- 6
Qwen2.5-VL-7B ↓2 27.9%
- 7
Qwen3-VL-2B-Thinking ↑8 25.7%
- 8
InternVL3.5-4B ↑3 19.7%
- 9
Qwen3-VL-2B-Instruct ↓4 19.5%
- 10
InternVL3.5-8B ↓2 19.3%
Diagrams & Tables
22 models we track have a score here.
- 1 GLM-4.1V-9B-Thinking ↑1 65.2%
- 2
Qwen2.5-VL-7B ↑2 61.9%
- 3
Qwen3-VL-4B-Instruct ↓2 55.4%
- 4
Qwen2.5-VL-3B ↑5 54.4%
- 5
Qwen2.5-Omni-7B ↑7 52.8%
- 6
InternVL3.5-8B ↑2 44.5%
- 7
InternVL3-9B 40.3%
- 8
InternVL2.5-4B ↑6 39.8%
- 9
InternVL3.5-4B ↑2 39.2%
- 10
Qwen3-VL-8B-Instruct ↓7 32.7%
General VQA
22 models we track have a score here.
- 1
Qwen3-VL-8B-Instruct ↑2 63.7%
- 2
Qwen3-VL-8B-Thinking ↑4 62.4%
- 3
Qwen3-VL-4B-Instruct ↓2 59.9%
- 4
Qwen3-VL-4B-Thinking ↑6 59.8%
- 5
Qwen2.5-VL-7B ↓1 56.8%
- 6 GLM-4.1V-9B-Thinking ↓4 54.1%
- 7
Qwen2.5-VL-3B ↑2 53.2%
- 8
Qwen3-VL-2B-Thinking ↑7 51.3%
- 9
InternVL3.5-8B ↓1 49.3%
- 10
InternVL3-9B ↓3 47.4%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The DatBench paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official DatBench 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
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 13 months ago, in October 2025, for Qwen3-VL-2B-Instruct. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the benchmark’s paper (26 entries).
- Pass Models scored 22 models we track, plus 4 other entries on its leaderboard.
- Warning Newest score The newest came 13 months ago, in October 2025, for Qwen3-VL-2B-Instruct.
- Note Frontier models None of its models are on the capabilities index.
- Pass Headroom The best is 64%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 22 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.