Benchmark Almanac

ColorBench

Gemini 2.0 Flash Feb 2025 59.6% leads since April 2025
Not in model cards

Models answer image questions about color perception, color reasoning, and robustness to color changes. The evaluator checks the selected answers against a single correct option.

The dataset contains 1,448 instances and more than 5,800 image-text questions across 11 tasks.

Released
April 2025
Kind
General visual questions
Level
Everyday
Scored
Checked automatically
Answers
Multiple choice
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
Strongest model tested
Gemini 2.0 Flash Feb 2025 #147 on the capability index
Last model added
February 2025
Scores refreshed
November 20th, 2026

In the authors’ words

Color plays an important role in human perception and usually provides critical clues in visual reasoning. However, it is unclear whether and how vision-language models (VLMs) can perceive, understand, and leverage color as humans. This paper introduces ColorBench, an innovative benchmark meticulously crafted to assess the capabilities of VLMs in color understanding, including color perception, reasoning, and robustness. By curating a suite of diverse test scenarios, with grounding in real applications, ColorBench evaluates how these models perceive colors, infer meanings from color-based cues, and maintain consistent performance under varying color transformations. Through an extensive evaluation of 32 VLMs with varying language models and vision encoders, our paper reveals some undiscovered findings: (i) The scaling law (larger models are better) still holds on ColorBench, while the language model plays a more important role than the vision encoder. (ii) However, the performance gaps across models are relatively small, indicating that color understanding has been largely neglected by existing VLMs. (iii) CoT reasoning improves color understanding accuracies and robustness, though they are vision-centric tasks. (iv) Color clues are indeed leveraged by VLMs on ColorBench but they can also mislead models in some tasks. These findings highlight the critical limitations of current VLMs and underscore the need to enhance color comprehension. Our ColorBenchcan serve as a foundational tool for advancing the study of human-level color understanding of multimodal AI.

Abstract of ColorBench: Can VLMs See and Understand the Colorful World? A Comprehensive Benchmark for Color Perception, Reasoning, and Robustness (arXiv:2504.10514, CC0)
32
Entries in its paper, 22 models we track

Gemini 2.0 Flash leads with 59.6% among models released by February 2025 Gemini 2.0 Flash leads with 59.6% 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. 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:

  1. 1 Gemini 2.0 Flash Feb 2025, CoT 59.6%
  2. 2 GPT-4o (snapshot unspecified) CoT 57.4%
  3. 3 LLaVA-OV-72B 51.9%
  4. 4 InternVL2.5-38B 50%
  5. 5 InternVL2_5-78B 48.8%
  6. 6 InternVL2.5-26B 46.8%
  7. 7 LLaVA-v1.6-34B 46.6%
  8. 8 InternVL2-26B 46.3%
  9. 9 Qwen2.5-VL-7B 46.2%
  10. 10 InternVL2-40B 45.6%
  11. 11 Cambrian-1-34B 45.3%
  12. 12 LLaVA-NeXT-72B 45.2%
  13. 13 InternVL2.5-8B 45.2%
  14. 14 LLaVA-OneVision 7B 44.7%
  15. 15 InternVL2-76B 44.6%
  16. 16 Eagle-X4-8B 44.1%
  17. 17 Eagle-X4-13B 43.7%
  18. 18 Eagle-X5-34B 43.4%
  19. 19 InternVL2-8B 43.1%
  20. 20 Cambrian-1-13B 42.8%
  21. 21 Cambrian-1-8B 42.3%
  22. 22 Cambrian-3B 41.5%
  23. 23 Qwen2.5-VL-3B 41.1%
  24. 24 Eagle-X5-7B 40%
  25. 25 InternVL2.5-2B 38.5%
  26. 26 InternVL2.5-1B 38.3%
  27. 27 llava-v1.6-vicuna-13b 36.4%
  28. 28 InternVL2-2B 36.4%
  29. 29 InternVL2-1B 33.6%
  30. 30 LLaVA-Next-m-7B 33.4%
  31. 31 LLaVA-OV-0.5B 32.6%
  32. 32 LLaVA-Next-v-7B 31.2%

Source: The ColorBench paper

#EntryScoreSettingDateSource
1 Gemini 2.0 Flash Feb 2025 59.6% CoT Feb 2025 Official leaderboard
2 GPT-4o (snapshot unspecified) closed weights 57.4% CoT Official leaderboard
3 LLaVA-OV-72B 51.9% Official leaderboard
4 InternVL2.5-38B 50% Dec 2024 Official leaderboard
5 InternVL2_5-78B 48.8% Dec 2024 Official leaderboard
6 InternVL2.5-26B 46.8% Dec 2024 Official leaderboard
7 LLaVA-v1.6-34B 46.6% Jan 2024 Official leaderboard
8 InternVL2-26B 46.3% Jul 2024 Official leaderboard
9 Qwen2.5-VL-7B 46.2% Jan 2025 Official leaderboard
10 InternVL2-40B 45.6% Jul 2024 Official leaderboard
11 Cambrian-1-34B 45.3% Jun 2024 Official leaderboard
12 LLaVA-NeXT-72B 45.2% May 2024 Official leaderboard
13 InternVL2.5-8B 45.2% Dec 2024 Official leaderboard
14 LLaVA-OneVision 7B 44.7% Aug 2024 Official leaderboard
15 InternVL2-76B 44.6% Official leaderboard
16 Eagle-X4-8B 44.1% Official leaderboard
17 Eagle-X4-13B 43.7% Official leaderboard
18 Eagle-X5-34B 43.4% Official leaderboard
19 InternVL2-8B 43.1% Jul 2024 Official leaderboard
20 Cambrian-1-13B 42.8% Jun 2024 Official leaderboard
21 Cambrian-1-8B 42.3% Jun 2024 Official leaderboard
22 Cambrian-3B 41.5% Official leaderboard
23 Qwen2.5-VL-3B 41.1% Jan 2025 Official leaderboard
24 Eagle-X5-7B 40% Aug 2024 Official leaderboard
25 InternVL2.5-2B 38.5% Dec 2024 Official leaderboard
26 InternVL2.5-1B 38.3% Dec 2024 Official leaderboard
27 llava-v1.6-vicuna-13b 36.4% Jan 2024 Official leaderboard
28 InternVL2-2B 36.4% Jul 2024 Official leaderboard
29 InternVL2-1B 33.6% Jul 2024 Official leaderboard
30 LLaVA-Next-m-7B 33.4% Official leaderboard
31 LLaVA-OV-0.5B 32.6% Official leaderboard
32 LLaVA-Next-v-7B 31.2% Official leaderboard
#ModelScoreSettingDateSource
1 Gemini 2.0 Flash Feb 2025 59.6% CoT Feb 2025 Official leaderboard
2 InternVL2.5-38B 50% Dec 2024 Official leaderboard
3 InternVL2_5-78B 48.8% Dec 2024 Official leaderboard
4 InternVL2.5-26B 46.8% Dec 2024 Official leaderboard
5 LLaVA-v1.6-34B 46.6% Jan 2024 Official leaderboard
6 InternVL2-26B 46.3% Jul 2024 Official leaderboard
7 Qwen2.5-VL-7B 46.2% Jan 2025 Official leaderboard
8 InternVL2-40B 45.6% Jul 2024 Official leaderboard
9 Cambrian-1-34B 45.3% Jun 2024 Official leaderboard
10 LLaVA-NeXT-72B 45.2% May 2024 Official leaderboard
11 InternVL2.5-8B 45.2% Dec 2024 Official leaderboard
12 LLaVA-OneVision 7B 44.7% Aug 2024 Official leaderboard
13 InternVL2-8B 43.1% Jul 2024 Official leaderboard
14 Cambrian-1-13B 42.8% Jun 2024 Official leaderboard
15 Cambrian-1-8B 42.3% Jun 2024 Official leaderboard
16 Qwen2.5-VL-3B 41.1% Jan 2025 Official leaderboard
17 Eagle-X5-7B 40% Aug 2024 Official leaderboard
18 InternVL2.5-2B 38.5% Dec 2024 Official leaderboard
19 InternVL2.5-1B 38.3% Dec 2024 Official leaderboard
20 llava-v1.6-vicuna-13b 36.4% Jan 2024 Official leaderboard
21 InternVL2-2B 36.4% Jul 2024 Official leaderboard
22 InternVL2-1B 33.6% Jul 2024 Official leaderboard

Rankings by part

We read 10 parts from the benchmark’s paper in 2 groups: color perception and color reasoning. Each ranks models on one part of the test. Gemini 2.0 Flash leads 6 of them, and 4 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

C’Recog

22 models we track have a score here.

  1. 1 Gemini 2.0 Flash Feb 2025 82.9%
  2. 2 InternVL2.5-8B ↑9 77.6%
  3. 3 Qwen2.5-VL-7B ↑4 76.3%
  4. 4 Cambrian-1-34B ↑5 75%
  5. 5 InternVL2_5-78B ↓2 75%
  6. 6 LLaVA-NeXT-72B ↑4 72.4%
  7. 7 Cambrian-1-8B ↑8 72.4%
  8. 8 InternVL2-26B ↓2 72.4%
  9. 9 InternVL2-8B ↑4 72.4%
  10. 10 InternVL2-40B ↓2 72.4%

Number one over time

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

Google DeepMind
Benchmark paper Not tested since Apr 2025

Source: The ColorBench 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% 2024 2025 Near the ceiling (85%) Gemini 2.0 Flash InternVL2.5-38B LLaVA-v1.6-34B 0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) Gemini 2.0 Flash InternVL2.5-38B

Source: Official ColorBench 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 21 months ago, in February 2025, for Gemini 2.0 Flash. 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 (32 entries).
  • Pass Models scored 22 models we track, plus 10 other entries on its leaderboard.
  • Warning Newest score The newest came 21 months ago, in February 2025, for Gemini 2.0 Flash.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 2.0 Flash (#147).
  • Pass Headroom The best is 59.6%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 23 rows for models we track and 10 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