Benchmark Almanac

Visual Reasoning Benchmark (Fab AI)

Gemini 3.7 Flash 89.9% leads since August 2026
Not in model cards

Models choose which picture completes a pattern, matches another shape, or shows a folded figure. Questions come from school exams in Zambia and India.

Fab AI’s leaderboard scores 67 model configurations. The authors describe 701 questions and publish example puzzles and scoring code. They keep the full exam images and answer keys private.

Made by
Fab AI
Released
September 2025
Kind
Spatial reasoning and counting
Level
School
Scored
Checked automatically
Answers
Multiple choice
Tasks
Public, plus a private test set
Measures
Capability
Language
English
In and out
Images, Text → Text
Interaction
Single turn
Items from
Exams and contests
Refresh
One fixed set
Run by
A host
In views
Vision
Strongest model tested
GPT-5.6 Sol #14 on the capability index
Last model added
August 2026
Scores refreshed
November 20th, 2026
59
Entries on its own leaderboard, 57 models we track

Gemini 3.7 Flash leads with 89.9% among models released by August 2026 Gemini 3.7 Flash leads with 89.9% among models released by August 2026

Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. 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 August 2026, 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 3.7 Flash 89.9%
  2. 2 Gemini 3.5 Flash 86%
  3. 3 Gemini 3.6 Flash 85.7%
  4. 4 GPT-5.6 Sol 84.9%
  5. 5 Gemini 3.1 Pro 81.3%
  6. 6 Gemini 3 Flash 78%
  7. 7 GPT-5.5 77.7%
  8. 8 GPT-5.4 medium 76.3%
  9. 9 Qwen3.5 397B-A17B 76%
  10. 10 Kimi K3 76%
  11. 11 Gemini 3 Pro 75.9%
  12. 12 Qwen 3.6 Plus 73.3%
  13. 13 Qwen3.5-Plus 72.6%
  14. 14 GPT-5.2 medium 67.3%
  15. 15 Qwen3.6 27B 66.2%
  16. 16 GPT-5.4 Mini medium 64.2%
  17. 17 Qwen 3.6 35B-A3B 64.2%
  18. 18 Qwen 3.6 Flash 63.5%
  19. 19 Qwen3.5-Flash 61.6%
  20. 20 Kimi K2.5 60.5%
  21. 21 Gemini 2.5 Pro Jun 2025 58.1%
  22. 22 Gemini 3.1 Flash-Lite 56.9%
  23. 23 GPT-5 medium 51.5%
  24. 24 Gemini 2.5 Flash Sep 2025 50.6%
  25. 25 GPT-5 mini medium 49.1%
  26. 26 GPT-5.1 medium 47.6%
  27. 27 Grok 4.20 reasoning 46.6%
  28. 28 o3 46.4%
  29. 29 Grok 4.3 46.2%
  30. 30 Qwen3-VL-30B-A3B-Thinking 45.4%
  31. 31 GLM-4.6V 44.5%
  32. 32 o4-mini 43.2%
  33. 33 GPT-4.1 mini 41.7%
  34. 34 Claude Opus 4 Thinking (low) 41.2%
  35. 35 Doubao-1.5 Vision Pro 39.7%
  36. 36 GPT-4.1 39.1%
  37. 37 Claude 3.7 Sonnet 38.7%
  38. 38 Grok 4 38.7%
  39. 39 Gemini 2.5 Flash-Lite Sep 2025 38.2%
  40. 40 Gemini 2.0 Flash Feb 2025 38.1%
  41. 41 GLM-4.1V-9B-Thinking 37.7%
  42. 42 Llama 4 Maverick 35.2%
  43. 43 GPT-5.4 Nano medium 35%
  44. 44 Claude Sonnet 4 34.7%
  45. 45 GPT-4o Mar 2025 33.2%
  46. 46 Pixtral Large 32.1%
  47. 47 Llama 4 Scout 31.2%
  48. 48 GPT-5 nano medium 31.1%
  49. 49 Qwen2.5-VL-72B 30%
  50. 50 Gemma 3 12B 29.2%
  51. 51 GPT-4o mini 28.8%
  52. 52 Mistral Medium 3 28.8%
  53. 53 Gemma 3 27B 28.7%
  54. 54 Grok 2 Vision (1212) 28.4%
  55. 55 GPT-4.1 nano 27.4%
  56. 56 Claude 3.5 Haiku 27.2%
  57. 57 Mistral Small 3.1 26.8%
  58. 58 LFM2.5 VL 1.6B 26%
  59. 59 Gemma 3 4B 23%

Source: Official Visual Reasoning Benchmark (Fab AI) leaderboard

Source: Official Visual Reasoning Benchmark (Fab AI) leaderboard

#EntryScoreSettingDateSource
1 Gemini 3.7 Flash 89.9% Aug 2026 Official leaderboard
2 Gemini 3.5 Flash 86% May 2026 Official leaderboard
3 Gemini 3.6 Flash 85.7% Jul 2026 Official leaderboard
4 GPT-5.6 Sol 84.9% Jul 2026 Official leaderboard
5 Gemini 3.1 Pro 81.3% Feb 2026 Official leaderboard
6 Gemini 3 Flash 78% Dec 2025 Official leaderboard
7 GPT-5.5 77.7% Apr 2026 Official leaderboard
8 GPT-5.4 76.3% medium Mar 2026 Official leaderboard
9 Qwen3.5 397B-A17B 76% Feb 2026 Official leaderboard
10 Kimi K3 76% Jul 2026 Official leaderboard
11 Gemini 3 Pro 75.9% Nov 2025 Official leaderboard
12 Qwen 3.6 Plus 73.3% Mar 2026 Official leaderboard
13 Qwen3.5-Plus 72.6% Feb 2026 Official leaderboard
14 GPT-5.2 67.3% medium Dec 2025 Official leaderboard
15 Qwen3.6 27B 66.2% Apr 2026 Official leaderboard
16 GPT-5.4 Mini 64.2% medium Mar 2026 Official leaderboard
17 Qwen 3.6 35B-A3B 64.2% Apr 2026 Official leaderboard
18 Qwen 3.6 Flash 63.5% Apr 2026 Official leaderboard
19 Qwen3.5-Flash 61.6% Feb 2026 Official leaderboard
20 Kimi K2.5 60.5% Jan 2026 Official leaderboard
21 Gemini 2.5 Pro Jun 2025 58.1% Jun 2025 Official leaderboard
22 Gemini 3.1 Flash-Lite 56.9% Mar 2026 Official leaderboard
23 GPT-5 51.5% medium Aug 2025 Official leaderboard
24 Gemini 2.5 Flash Sep 2025 50.6% Sep 2025 Official leaderboard
25 GPT-5 mini 49.1% medium Aug 2025 Official leaderboard
26 GPT-5.1 47.6% medium Nov 2025 Official leaderboard
27 Grok 4.20 46.6% reasoning Feb 2026 Official leaderboard
28 o3 46.4% Apr 2025 Official leaderboard
29 Grok 4.3 46.2% Apr 2026 Official leaderboard
30 Qwen3-VL-30B-A3B-Thinking 45.4% Oct 2025 Official leaderboard
31 GLM-4.6V 44.5% Dec 2025 Official leaderboard
32 o4-mini 43.2% Apr 2025 Official leaderboard
33 GPT-4.1 mini 41.7% Apr 2025 Official leaderboard
34 Claude Opus 4 41.2% Thinking (low) May 2025 Official leaderboard
35 Doubao-1.5 Vision Pro 39.7% Official leaderboard
36 GPT-4.1 39.1% Apr 2025 Official leaderboard
37 Claude 3.7 Sonnet 38.7% Feb 2025 Official leaderboard
38 Grok 4 38.7% Jul 2025 Official leaderboard
39 Gemini 2.5 Flash-Lite Sep 2025 38.2% Sep 2025 Official leaderboard
40 Gemini 2.0 Flash Feb 2025 38.1% Feb 2025 Official leaderboard
41 GLM-4.1V-9B-Thinking 37.7% Jul 2025 Official leaderboard
42 Llama 4 Maverick 35.2% Apr 2025 Official leaderboard
43 GPT-5.4 Nano 35% medium Mar 2026 Official leaderboard
44 Claude Sonnet 4 34.7% May 2025 Official leaderboard
45 GPT-4o Mar 2025 33.2% Mar 2025 Official leaderboard
46 Pixtral Large 32.1% Nov 2024 Official leaderboard
47 Llama 4 Scout 31.2% Apr 2025 Official leaderboard
48 GPT-5 nano 31.1% medium Aug 2025 Official leaderboard
49 Qwen2.5-VL-72B 30% Jan 2025 Official leaderboard
50 Gemma 3 12B 29.2% Mar 2025 Official leaderboard
51 GPT-4o mini 28.8% Jul 2024 Official leaderboard
52 Mistral Medium 3 28.8% May 2025 Official leaderboard
53 Gemma 3 27B 28.7% Mar 2025 Official leaderboard
54 Grok 2 Vision (1212) 28.4% Dec 2024 Official leaderboard
55 GPT-4.1 nano 27.4% Apr 2025 Official leaderboard
56 Claude 3.5 Haiku 27.2% Oct 2024 Official leaderboard
57 Mistral Small 3.1 26.8% Mar 2025 Official leaderboard
58 LFM2.5 VL 1.6B 26% Official leaderboard
59 Gemma 3 4B 23% Mar 2025 Official leaderboard
#ModelScoreSettingDateSource
1 Gemini 3.7 Flash 89.9% Aug 2026 Official leaderboard
2 Gemini 3.5 Flash 86% May 2026 Official leaderboard
3 Gemini 3.6 Flash 85.7% Jul 2026 Official leaderboard
4 GPT-5.6 Sol 84.9% Jul 2026 Official leaderboard
5 Gemini 3.1 Pro 81.3% Feb 2026 Official leaderboard
6 Gemini 3 Flash 78% Dec 2025 Official leaderboard
7 GPT-5.5 77.7% Apr 2026 Official leaderboard
8 GPT-5.4 76.3% medium Mar 2026 Official leaderboard
9 Qwen3.5 397B-A17B 76% Feb 2026 Official leaderboard
10 Kimi K3 76% Jul 2026 Official leaderboard
11 Gemini 3 Pro 75.9% Nov 2025 Official leaderboard
12 Qwen 3.6 Plus 73.3% Mar 2026 Official leaderboard
13 Qwen3.5-Plus 72.6% Feb 2026 Official leaderboard
14 GPT-5.2 67.3% medium Dec 2025 Official leaderboard
15 Qwen3.6 27B 66.2% Apr 2026 Official leaderboard
16 GPT-5.4 Mini 64.2% medium Mar 2026 Official leaderboard
17 Qwen 3.6 35B-A3B 64.2% Apr 2026 Official leaderboard
18 Qwen 3.6 Flash 63.5% Apr 2026 Official leaderboard
19 Qwen3.5-Flash 61.6% Feb 2026 Official leaderboard
20 Kimi K2.5 60.5% Jan 2026 Official leaderboard
21 Gemini 2.5 Pro Jun 2025 58.1% Jun 2025 Official leaderboard
22 Gemini 3.1 Flash-Lite 56.9% Mar 2026 Official leaderboard
23 GPT-5 51.5% medium Aug 2025 Official leaderboard
24 Gemini 2.5 Flash Sep 2025 50.6% Sep 2025 Official leaderboard
25 GPT-5 mini 49.1% medium Aug 2025 Official leaderboard
26 GPT-5.1 47.6% medium Nov 2025 Official leaderboard
27 Grok 4.20 46.6% reasoning Feb 2026 Official leaderboard
28 o3 46.4% Apr 2025 Official leaderboard
29 Grok 4.3 46.2% Apr 2026 Official leaderboard
30 Qwen3-VL-30B-A3B-Thinking 45.4% Oct 2025 Official leaderboard
31 GLM-4.6V 44.5% Dec 2025 Official leaderboard
32 o4-mini 43.2% Apr 2025 Official leaderboard
33 GPT-4.1 mini 41.7% Apr 2025 Official leaderboard
34 Claude Opus 4 41.2% Thinking (low) May 2025 Official leaderboard
35 GPT-4.1 39.1% Apr 2025 Official leaderboard
36 Claude 3.7 Sonnet 38.7% Feb 2025 Official leaderboard
37 Grok 4 38.7% Jul 2025 Official leaderboard
38 Gemini 2.5 Flash-Lite Sep 2025 38.2% Sep 2025 Official leaderboard
39 Gemini 2.0 Flash Feb 2025 38.1% Feb 2025 Official leaderboard
40 GLM-4.1V-9B-Thinking 37.7% Jul 2025 Official leaderboard
41 Llama 4 Maverick 35.2% Apr 2025 Official leaderboard
42 GPT-5.4 Nano 35% medium Mar 2026 Official leaderboard
43 Claude Sonnet 4 34.7% May 2025 Official leaderboard
44 GPT-4o Mar 2025 33.2% Mar 2025 Official leaderboard
45 Pixtral Large 32.1% Nov 2024 Official leaderboard
46 Llama 4 Scout 31.2% Apr 2025 Official leaderboard
47 GPT-5 nano 31.1% medium Aug 2025 Official leaderboard
48 Qwen2.5-VL-72B 30% Jan 2025 Official leaderboard
49 Gemma 3 12B 29.2% Mar 2025 Official leaderboard
50 GPT-4o mini 28.8% Jul 2024 Official leaderboard
51 Mistral Medium 3 28.8% May 2025 Official leaderboard
52 Gemma 3 27B 28.7% Mar 2025 Official leaderboard
53 Grok 2 Vision (1212) 28.4% Dec 2024 Official leaderboard
54 GPT-4.1 nano 27.4% Apr 2025 Official leaderboard
55 Claude 3.5 Haiku 27.2% Oct 2024 Official leaderboard
56 Mistral Small 3.1 26.8% Mar 2025 Official leaderboard
57 Gemma 3 4B 23% Mar 2025 Official leaderboard

Rankings by part

We read 6 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as match (figure), match (process) and odd one out. Gemini 3.7 Flash leads 3 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

match (figure)

57 models we track have a score here.

  1. 1 Gemini 3.7 Flash 87.4%
  2. 2 Gemini 3.5 Flash 85.2%
  3. 3 Gemini 3.6 Flash 84.7%
  4. 4 Gemini 3.1 Pro ↑1 82%
  5. 5 GPT-5.6 Sol ↓1 77%
  6. 6 Gemini 3 Flash 76%
  7. 7 GPT-5.5 74.3%
  8. 8 Gemini 3 Pro ↑3 73.8%
  9. 9 Qwen3.5 397B-A17B 69.9%
  10. 10 GPT-5.4 ↓2 69.4%

More effort adds a median 14 points

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
GPT-5.4
medium
76.3%
GPT-5.2
none medium
67.3%
GPT-5
minimal medium
51.5%
GPT-5 mini
minimal medium
49.1%
Claude Opus 4
Thinking (l…
41.2%
Claude Sonnet 4
Thinking (l…
34.7%
GPT-5 nano
minimal medium
31.1%

Source: Official Visual Reasoning Benchmark (Fab AI) leaderboard

More effort adds a median 14 points

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
GPT-5.4
medium
76.3%
GPT-5.2
none medium
67.3%
GPT-5
minimal medium
51.5%
GPT-5 mini
minimal medium
49.1%
Claude Opus 4
Thinking (l…
41.2%
Claude Sonnet 4
Thinking (l…
34.7%
GPT-5 nano
minimal medium
31.1%

Source: Official Visual Reasoning Benchmark (Fab AI) leaderboard

6 models have held number one on the official leaderboard since its release in September 2025

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

Google DeepMind
Own leaderboard Passed 85% in May 2026, by Gemini 3.5 Flash

Source: Official Visual Reasoning Benchmark (Fab AI) leaderboard

Gemini 3.5 Flash on the official leaderboard passed 85% in May 2026

That was 8 months after its release. By the same age, 16% of the benchmarks from 2025 had passed 85%. All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 3.7 Flash Gemini 3.1 Pro Gemini 3 Pro Gemini 2.5 Pro o3 GPT-4.1 mini Pixtral Large 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 3.7 Flash Gemini 3.1 Pro

Source: Official Visual Reasoning Benchmark (Fab AI) 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

All 9 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access Public tasks, plus a private test set.
  • Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
  • Pass Independent runs Scores from the official leaderboard (59 entries).
  • Pass Models scored 57 models we track, plus 2 other entries on its leaderboard.
  • Pass Newest score The newest is from August 2026, for Gemini 3.7 Flash.
  • Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is GPT-5.6 Sol (#14).
  • Note Headroom Saturated. The best is 89.9%, past 85% since May 2026.
Leaderboard import 3 pass
  • Pass Importer We read the board on fab-ai.org: 65 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