Benchmark Almanac

Argus Inspection

Gemini 2.5 Flash May 2025 0.44 leads since June 2025
Not in model cards

Models inspect images for small details that change the sensible response to a question. A judge scores whether the response identifies the relevant detail and chooses an appropriate action. The score combines those judgments and assigns zero when a separate description fails to recognize the detail.

The benchmark contains 1,430 entries at basic and deceptive difficulty levels. GPT-4o judges two four-point dimensions, which a nonlinear function combines with a visual-recognition gate. Scores depend on the published judge and function parameters.

Released
June 2025
Kind
General visual questions
Level
Everyday
Scored
Graded by a model
Answers
Open-ended
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
o3 #84 on the capability index
Last model added
May 2025
Scores refreshed
November 20th, 2026

In the authors’ words

As Multimodal Large Language Models (MLLMs) continue to evolve, their cognitive and reasoning capabilities have seen remarkable progress. However, challenges in visual fine-grained perception and commonsense causal inference persist. This paper introduces Argus Inspection, a multimodal benchmark with two levels of difficulty, emphasizing detailed visual recognition while incorporating real-world commonsense understanding to evaluate causal reasoning abilities. Expanding on it, we present the Eye of Panoptes framework, which integrates a binary parametric Sigmoid metric with an indicator function, enabling a more holistic evaluation of MLLMs' responses in opinion-based reasoning tasks. Experiments conducted on 26 mainstream MLLMs reveal that the highest performance in visual fine-grained reasoning reaches only 0.46, highlighting considerable potential for enhancement. Our research offers valuable perspectives for the continued refinement of MLLMs.

Abstract of Argus Inspection: Do Multimodal Large Language Models Possess the Eye of Panoptes? (arXiv:2506.14805, CC0)
26
Entries in its paper, 24 models we track

Gemini 2.5 Flash leads with 0.44 among models released by May 2025 Gemini 2.5 Flash leads with 0.44 among models released by May 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 authors give separate overall scores for basic and deceptive questions, so this is their plain mean. 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 May 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 2.5 Flash May 2025 0.44
  2. 2 GPT-4.1 0.43
  3. 3 Gemini 2.5 Pro May 2025 0.41
  4. 4 o3 0.34
  5. 5 GPT-4.5 0.33
  6. 6 Seed1.5-VL 0.33
  7. 7 Doubao-1.5-Vision-Pro-250328 0.32
  8. 8 GPT-4o Nov 2024 0.31
  9. 9 Gemini-1.5-Pro (snapshot unspecified) 0.31
  10. 10 Gemini 2.0 Flash Feb 2025 0.28
  11. 11 Claude 3.5 Sonnet 0.28
  12. 12 Qwen2.5-VL-72B 0.28
  13. 13 Claude 3.7 Sonnet 0.27
  14. 14 o4-mini 0.26
  15. 15 Claude Sonnet 4 0.23
  16. 16 Qwen2.5-VL-32B 0.23
  17. 17 Grok 2 Vision (1212) 0.23
  18. 18 QVQ-72B-Preview 0.21
  19. 19 Llama 4 Maverick 0.18
  20. 20 Llama 4 Scout 0.18
  21. 21 InternVL2_5-78B 0.18
  22. 22 o1 0.17
  23. 23 Llama 3.2 90B 0.12
  24. 24 Llama 3.2 11B 0.09
  25. 25 LLaVA-v1.6-34B 0.09
  26. 26 DeepSeek-VL2 0.07

Source: The Argus Inspection paper

#EntryScoreSettingDateSource
1 Gemini 2.5 Flash May 2025 0.44 May 2025 Official leaderboard
2 GPT-4.1 0.43 Apr 2025 Official leaderboard
3 Gemini 2.5 Pro May 2025 0.41 May 2025 Official leaderboard
4 o3 0.34 Apr 2025 Official leaderboard
5 GPT-4.5 0.33 Feb 2025 Official leaderboard
6 Seed1.5-VL 0.33 May 2025 Official leaderboard
7 Doubao-1.5-Vision-Pro-250328 0.32 Official leaderboard
8 GPT-4o Nov 2024 0.31 Nov 2024 Official leaderboard
9 Gemini-1.5-Pro (snapshot unspecified) 0.31 Official leaderboard
10 Gemini 2.0 Flash Feb 2025 0.28 Feb 2025 Official leaderboard
11 Claude 3.5 Sonnet 0.28 Jun 2024 Official leaderboard
12 Qwen2.5-VL-72B 0.28 Jan 2025 Official leaderboard
13 Claude 3.7 Sonnet 0.27 Feb 2025 Official leaderboard
14 o4-mini 0.26 Apr 2025 Official leaderboard
15 Claude Sonnet 4 0.23 May 2025 Official leaderboard
16 Qwen2.5-VL-32B 0.23 Mar 2025 Official leaderboard
17 Grok 2 Vision (1212) 0.23 Dec 2024 Official leaderboard
18 QVQ-72B-Preview 0.21 Dec 2024 Official leaderboard
19 Llama 4 Maverick 0.18 Apr 2025 Official leaderboard
20 Llama 4 Scout 0.18 Apr 2025 Official leaderboard
21 InternVL2_5-78B 0.18 Dec 2024 Official leaderboard
22 o1 0.17 Dec 2024 Official leaderboard
23 Llama 3.2 90B 0.12 Sep 2024 Official leaderboard
24 Llama 3.2 11B 0.09 Sep 2024 Official leaderboard
25 LLaVA-v1.6-34B 0.09 Jan 2024 Official leaderboard
26 DeepSeek-VL2 0.07 Dec 2024 Official leaderboard
#ModelScoreSettingDateSource
1 Gemini 2.5 Flash May 2025 0.44 May 2025 Official leaderboard
2 GPT-4.1 0.43 Apr 2025 Official leaderboard
3 Gemini 2.5 Pro May 2025 0.41 May 2025 Official leaderboard
4 o3 0.34 Apr 2025 Official leaderboard
5 GPT-4.5 0.33 Feb 2025 Official leaderboard
6 Seed1.5-VL 0.33 May 2025 Official leaderboard
7 GPT-4o Nov 2024 0.31 Nov 2024 Official leaderboard
8 Gemini 2.0 Flash Feb 2025 0.28 Feb 2025 Official leaderboard
9 Claude 3.5 Sonnet 0.28 Jun 2024 Official leaderboard
10 Qwen2.5-VL-72B 0.28 Jan 2025 Official leaderboard
11 Claude 3.7 Sonnet 0.27 Feb 2025 Official leaderboard
12 o4-mini 0.26 Apr 2025 Official leaderboard
13 Claude Sonnet 4 0.23 May 2025 Official leaderboard
14 Qwen2.5-VL-32B 0.23 Mar 2025 Official leaderboard
15 Grok 2 Vision (1212) 0.23 Dec 2024 Official leaderboard
16 QVQ-72B-Preview 0.21 Dec 2024 Official leaderboard
17 Llama 4 Maverick 0.18 Apr 2025 Official leaderboard
18 Llama 4 Scout 0.18 Apr 2025 Official leaderboard
19 InternVL2_5-78B 0.18 Dec 2024 Official leaderboard
20 o1 0.17 Dec 2024 Official leaderboard
21 Llama 3.2 90B 0.12 Sep 2024 Official leaderboard
22 Llama 3.2 11B 0.09 Sep 2024 Official leaderboard
23 LLaVA-v1.6-34B 0.09 Jan 2024 Official leaderboard
24 DeepSeek-VL2 0.07 Dec 2024 Official leaderboard

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 Jun 2025

Source: The Argus Inspection 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
0.00 0.10 0.20 0.30 0.40 0.50 2024 2025 0.00 0.10 0.20 0.30 0.40 0.50 2024 2025

Source: Official Argus Inspection 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

8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 18 months ago, in May 2025, for Claude Sonnet 4. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
  • Pass Independent runs Scores from the benchmark’s paper (26 entries).
  • Pass Models scored 24 models we track, plus 2 other entries on its leaderboard.
  • Warning Newest score The newest came 18 months ago, in May 2025, for Claude Sonnet 4.
  • Note Frontier models None of the top 10 has a score. The strongest tested is o3 (#84).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 24 rows for models we track and 2 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