Benchmark Almanac

DatBench

Qwen3-VL-4B-Instruct 64% leads since January 2026
Not in model cards

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
In and out
Text, Images → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
In views
Vision
Last model added
October 2025
Scores refreshed
November 20th, 2026
26
Entries in its paper, 22 models we track

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:

  1. 1 Qwen3-VL-4B-Instruct 64%
  2. 2 GLM-4.1V-9B-Base 58.7%
  3. 3 GLM-4.1V-9B-Thinking 58.3%
  4. 4 Qwen3-VL-8B-Instruct 58.1%
  5. 5 R-4B 54.7%
  6. 6 Qwen2.5-VL-7B 53.2%
  7. 7 Qwen3-VL-2B-Instruct 52.3%
  8. 8 Qwen3-VL-8B-Thinking 51.6%
  9. 9 InternVL3-9B 51.2%
  10. 10 InternVL3.5-8B 49.7%
  11. 11 Qwen2.5-VL-3B 49.6%
  12. 12 Qwen3-VL-4B-Thinking 48.7%
  13. 13 InternVL3.5-4B 47.1%
  14. 14 Qwen2.5-Omni-7B 47%
  15. 15 InternVL2.5-8B 45.5%
  16. 16 InternVL2.5-4B 44.8%
  17. 17 Qwen3-VL-2B-Thinking 43.9%
  18. 18 InternVL2-8B 41.2%
  19. 19 InternVL3-2B 39.5%
  20. 20 Qwen2.5-Omni-3B 38.5%
  21. 21 InternVL3_5-2B-Instruct 36.6%
  22. 22 InternVL2-4B 34.3%
  23. 23 InternVL2.5-2B 32.9%
  24. 24 Phi-3.5-vision-instruct 31.4%
  25. 25 InternVL2-2B 25.1%
  26. 26 Gemma 3 4B 15%

Source: The DatBench paper

#EntryScoreSettingDateSource
1 Qwen3-VL-4B-Instruct 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 Qwen3-VL-8B-Instruct 58.1% Oct 2025 Official leaderboard
5 R-4B 54.7% Official leaderboard
6 Qwen2.5-VL-7B 53.2% Jan 2025 Official leaderboard
7 Qwen3-VL-2B-Instruct 52.3% Oct 2025 Official leaderboard
8 Qwen3-VL-8B-Thinking 51.6% Oct 2025 Official leaderboard
9 InternVL3-9B 51.2% Apr 2025 Official leaderboard
10 InternVL3.5-8B 49.7% Aug 2025 Official leaderboard
11 Qwen2.5-VL-3B 49.6% Jan 2025 Official leaderboard
12 Qwen3-VL-4B-Thinking 48.7% Oct 2025 Official leaderboard
13 InternVL3.5-4B 47.1% Aug 2025 Official leaderboard
14 Qwen2.5-Omni-7B 47% Mar 2025 Official leaderboard
15 InternVL2.5-8B 45.5% Dec 2024 Official leaderboard
16 InternVL2.5-4B 44.8% Dec 2024 Official leaderboard
17 Qwen3-VL-2B-Thinking 43.9% Oct 2025 Official leaderboard
18 InternVL2-8B 41.2% Jul 2024 Official leaderboard
19 InternVL3-2B 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 InternVL2-4B 34.3% Jul 2024 Official leaderboard
23 InternVL2.5-2B 32.9% Dec 2024 Official leaderboard
24 Phi-3.5-vision-instruct 31.4% Aug 2024 Official leaderboard
25 InternVL2-2B 25.1% Jul 2024 Official leaderboard
26 Gemma 3 4B 15% Mar 2025 Official leaderboard
#ModelScoreSettingDateSource
1 Qwen3-VL-4B-Instruct 64% Oct 2025 Official leaderboard
2 GLM-4.1V-9B-Thinking 58.3% Jul 2025 Official leaderboard
3 Qwen3-VL-8B-Instruct 58.1% Oct 2025 Official leaderboard
4 Qwen2.5-VL-7B 53.2% Jan 2025 Official leaderboard
5 Qwen3-VL-2B-Instruct 52.3% Oct 2025 Official leaderboard
6 Qwen3-VL-8B-Thinking 51.6% Oct 2025 Official leaderboard
7 InternVL3-9B 51.2% Apr 2025 Official leaderboard
8 InternVL3.5-8B 49.7% Aug 2025 Official leaderboard
9 Qwen2.5-VL-3B 49.6% Jan 2025 Official leaderboard
10 Qwen3-VL-4B-Thinking 48.7% Oct 2025 Official leaderboard
11 InternVL3.5-4B 47.1% Aug 2025 Official leaderboard
12 Qwen2.5-Omni-7B 47% Mar 2025 Official leaderboard
13 InternVL2.5-8B 45.5% Dec 2024 Official leaderboard
14 InternVL2.5-4B 44.8% Dec 2024 Official leaderboard
15 Qwen3-VL-2B-Thinking 43.9% Oct 2025 Official leaderboard
16 InternVL2-8B 41.2% Jul 2024 Official leaderboard
17 InternVL3-2B 39.5% Apr 2025 Official leaderboard
18 InternVL2-4B 34.3% Jul 2024 Official leaderboard
19 InternVL2.5-2B 32.9% Dec 2024 Official leaderboard
20 Phi-3.5-vision-instruct 31.4% Aug 2024 Official leaderboard
21 InternVL2-2B 25.1% Jul 2024 Official leaderboard
22 Gemma 3 4B 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. 1 GLM-4.1V-9B-Thinking ↑1 76%
  2. 2 Qwen3-VL-8B-Thinking ↑4 73.2%
  3. 3 Qwen3-VL-8B-Instruct 70.5%
  4. 4 Qwen3-VL-4B-Thinking ↑6 68.9%
  5. 5 Qwen2.5-VL-7B ↓1 64.1%
  6. 6 Qwen3-VL-4B-Instruct ↓5 63.8%
  7. 7 Qwen3-VL-2B-Thinking ↑8 62.3%
  8. 8 InternVL3.5-8B 60.4%
  9. 9 InternVL3.5-4B ↑2 58.8%
  10. 10 InternVL3-9B ↓3 54.8%

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Qwen
Benchmark paper Not tested since Jan 2026

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.

Plot with other benchmarks
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Qwen3-VL-4B-Instruct GLM-4.1V-9B-Thinking Qwen2.5-VL-7B InternVL2.5-8B InternVL2-8B 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Qwen3-VL-4B-Instruct Qwen2.5-VL-7B

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

All 636 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.

How we check