Benchmark Almanac

CAIS Vision Capabilities Index

Gemini 4 Argon 83.3 leads since Sep 2026 Not in model cards

Averages model scores on five vision tests run by the Center for AI Safety: ERQA, MindCube, SpatialViz-Bench, IntPhys 2 and EnigmaEval.

One of three CAIS dashboard indexes, next to the Text Capabilities Index and the Risk Index. CAIS announced the dashboard in AI Safety Newsletter 66 on 2025-12-02. 52 models had scores on 2026-10-23.

Made by Center for AI Safety Released Dec 2025 Composite indices All tasks public Last model added Oct 2026 Scores refreshed 2 Nov 2026
Level
Research frontier Researchers, on open problems or tasks that take days
Scored
Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
Answers
Short answer A number, word or short string matched to a key.
centerforaisafety/simple-evals 23 stars Leaderboard dashboard.safe.ai
53
Models scored on its own leaderboard

Gemini 4 Argon leads with 83.3

Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best.

53 models. Scroll sideways for the rest.

  1. 83.3 Gemini 4 Argon
  2. 83.0 GPT-6 Astra
  3. 77.9 GPT-6.1 Sol
  4. 76.8 Claude Opus 5.5
  5. 71.2 Muse Wonder
  6. 69.9 Gemini 3.8 Flash
  7. 68.9 Claude Fable 5.1
  8. 68.7 Gemini 3.7 Flash
  9. 68.0 Claude Opus 5
  10. 66.6 Claude Fable 5
  11. 66.4 GPT-5.6 Sol
  12. 64.6 Gemini 3.5 Flash
  13. 64.3 GPT-5.5
  14. 62.1 Gemini 3.1 Pro
  15. 60.6 Kimi K3
  16. 60.1 Muse Spark 1.1
  17. 59.4 Muse Spark Open
  18. 59.3 Gemini 3 Flash
  19. 57.7 GPT-5.4
  20. 57.1 Gemini 3 Pro
  21. 56.2 Grok 4.5
  22. 55.3 Kimi K2.6
  23. 55.1 GPT-5.6 Terra
  24. 53.7 GPT-5.6 Luna
  25. 53.0 Claude Opus 4.8
  26. 52.2 GPT-5.2
  27. 51.2 Claude Opus 4.7
  28. 48.5 Grok 4.20
  29. 48.4 Grok 4.3
  30. 47.8 GPT-5
  31. 47.7 Claude Opus 4.6
  32. 47.5 GPT-5 mini
  33. 47.4 GPT-5.1
  34. 46.6 GPT-5.4 Mini
  35. 46.5 o3
  36. 46.3 Claude Sonnet 4.6
  37. 45.7 Gemini 2.5 Pro
  38. 45.4 Claude Opus 4.5
  39. 44.7 Gemini 3.1 Flash-Lite
  40. 42.5 Grok 4
  41. 42.1 Claude Sonnet 4.5
  42. 41.1 Gemini 2.5 Flash
  43. 40.9 o1
  44. 39.5 Grok 4.1 Fast
  45. 38.9 GPT-5.4 Nano
  46. 38.1 Gemini 2.5 Flash-Lite
  47. 37.7 Grok 4 Fast
  48. 37.5 Claude Haiku 4.5
  49. 36.8 o4-mini
  50. 36.0 GPT-4o
  51. 34.3 GPT-5 nano
  52. 33.9 Claude 3.7 Sonnet
  53. 26.4 Claude Sonnet 4
  1. 1 Gemini 4 Argon high 83.3
  2. 2 GPT-6 Astra high 83.0
  3. 3 GPT-6.1 Sol high 77.9
  4. 4 Claude Opus 5.5 high 76.8
  5. 5 Muse Wonder high 71.2
  6. 6 Gemini 3.8 Flash high 69.9
  7. 7 Claude Fable 5.1 high 68.9
  8. 8 Gemini 3.7 Flash high 68.7
  9. 9 Claude Opus 5 high 68.0
  10. 10 Claude Fable 5 high 66.6
  11. 11 GPT-5.6 Sol high 66.4
  12. 12 Gemini 3.5 Flash 64.6
  13. 13 GPT-5.5 high 64.3
  14. 14 Gemini 3.1 Pro high 62.1
  15. 15 Kimi K3 high 60.6
  16. 16 Muse Spark 1.1 high 60.1
  17. 17 Muse Spark Open high 59.4
  18. 18 Gemini 3 Flash high 59.3
  19. 19 GPT-5.4 high 57.7
  20. 20 Gemini 3 Pro high 57.1
  21. 21 Grok 4.5 56.2
  22. 22 Kimi K2.6 55.3
  23. 23 GPT-5.6 Terra high 55.1
  24. 24 GPT-5.6 Luna high 53.7
  25. 25 Claude Opus 4.8 high 53.0
  26. 26 GPT-5.2 high 52.2
  27. 27 Claude Opus 4.7 51.2
  28. 28 Grok 4.20 48.5
  29. 29 Grok 4.3 48.4
  30. 30 GPT-5 high 47.8
  31. 31 Claude Opus 4.6 47.7
  32. 32 GPT-5 mini high 47.5
  33. 33 GPT-5.1 high 47.4
  34. 34 GPT-5.4 Mini high 46.6
  35. 35 o3 high 46.5
  36. 36 Claude Sonnet 4.6 46.3
  37. 37 Gemini 2.5 Pro Jun 2025, high 45.7
  38. 38 Claude Opus 4.5 thinking 45.4
  39. 39 Gemini 3.1 Flash-Lite high 44.7
  40. 40 Grok 4 42.5
  41. 41 Claude Sonnet 4.5 thinking 42.1
  42. 42 Gemini 2.5 Flash Jun 2025, high 41.1
  43. 43 o1 40.9
  44. 44 Grok 4.1 Fast 39.5
  45. 45 GPT-5.4 Nano high 38.9
  46. 46 Gemini 2.5 Flash-Lite Jun 2025, high 38.1
  47. 47 Grok 4 Fast 37.7
  48. 48 Claude Haiku 4.5 thinking 37.5
  49. 49 o4-mini high 36.8
  50. 50 GPT-4o Nov 2024 36.0
  51. 51 GPT-5 nano high 34.3
  52. 52 Claude 3.7 Sonnet thinking 33.9
  53. 53 Claude Sonnet 4 thinking 26.4
#ModelScoreSettingDateSource
1 Gemini 4 Argon 83.3 high Sep 2026 Official leaderboard
2 GPT-6 Astra 83.0 high Sep 2026 Official leaderboard
3 GPT-6.1 Sol 77.9 high Sep 2026 Official leaderboard
4 Claude Opus 5.5 76.8 high Sep 2026 Official leaderboard
5 Muse Wonder 71.2 high Oct 2026 Official leaderboard
6 Gemini 3.8 Flash 69.9 high Sep 2026 Official leaderboard
7 Claude Fable 5.1 68.9 high Sep 2026 Official leaderboard
8 Gemini 3.7 Flash 68.7 high Aug 2026 Official leaderboard
9 Claude Opus 5 68.0 high Jul 2026 Official leaderboard
10 Claude Fable 5 66.6 high Jun 2026 Official leaderboard
11 GPT-5.6 Sol 66.4 high Jul 2026 Official leaderboard
12 Gemini 3.5 Flash 64.6 May 2026 Official leaderboard
13 GPT-5.5 64.3 high Apr 2026 Official leaderboard
14 Gemini 3.1 Pro 62.1 high Feb 2026 Official leaderboard
15 Kimi K3 60.6 high Jul 2026 Official leaderboard
16 Muse Spark 1.1 60.1 high Jul 2026 Official leaderboard
17 Muse Spark Open 59.4 high Oct 2026 Official leaderboard
18 Gemini 3 Flash 59.3 high Dec 2025 Official leaderboard
19 GPT-5.4 57.7 high Mar 2026 Official leaderboard
20 Gemini 3 Pro 57.1 high Nov 2025 Official leaderboard
21 Grok 4.5 56.2 Jul 2026 Official leaderboard
22 Kimi K2.6 55.3 Apr 2026 Official leaderboard
23 GPT-5.6 Terra 55.1 high Jul 2026 Official leaderboard
24 GPT-5.6 Luna 53.7 high Jul 2026 Official leaderboard
25 Claude Opus 4.8 53.0 high May 2026 Official leaderboard
26 GPT-5.2 52.2 high Dec 2025 Official leaderboard
27 Claude Opus 4.7 51.2 Apr 2026 Official leaderboard
28 Grok 4.20 48.5 Feb 2026 Official leaderboard
29 Grok 4.3 48.4 Apr 2026 Official leaderboard
30 GPT-5 47.8 high Aug 2025 Official leaderboard
31 Claude Opus 4.6 47.7 Feb 2026 Official leaderboard
32 GPT-5 mini 47.5 high Aug 2025 Official leaderboard
33 GPT-5.1 47.4 high Nov 2025 Official leaderboard
34 GPT-5.4 Mini 46.6 high Mar 2026 Official leaderboard
35 o3 46.5 high Apr 2025 Official leaderboard
36 Claude Sonnet 4.6 46.3 Feb 2026 Official leaderboard
37 Gemini 2.5 Pro Jun 2025 45.7 high Jun 2025 Official leaderboard
38 Claude Opus 4.5 45.4 thinking Nov 2025 Official leaderboard
39 Gemini 3.1 Flash-Lite 44.7 high Mar 2026 Official leaderboard
40 Grok 4 42.5 Jul 2025 Official leaderboard
41 Claude Sonnet 4.5 42.1 thinking Sep 2025 Official leaderboard
42 Gemini 2.5 Flash Jun 2025 41.1 high Jun 2025 Official leaderboard
43 o1 40.9 Dec 2024 Official leaderboard
44 Grok 4.1 Fast 39.5 Nov 2025 Official leaderboard
45 GPT-5.4 Nano 38.9 high Mar 2026 Official leaderboard
46 Gemini 2.5 Flash-Lite Jun 2025 38.1 high Jun 2025 Official leaderboard
47 Grok 4 Fast 37.7 Sep 2025 Official leaderboard
48 Claude Haiku 4.5 37.5 thinking Oct 2025 Official leaderboard
49 o4-mini 36.8 high Apr 2025 Official leaderboard
50 GPT-4o Nov 2024 36.0 Nov 2024 Official leaderboard
51 GPT-5 nano 34.3 high Aug 2025 Official leaderboard
52 Claude 3.7 Sonnet 33.9 thinking Feb 2025 Official leaderboard
53 Claude Sonnet 4 26.4 thinking May 2025 Official leaderboard

12 models have held number one since its release in Dec 2025

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

Google DeepMind OpenAI Anthropic
Own leaderboard 12 leaders
Gemini 4 Argon 83.3

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard
20.0 40.0 60.0 80.0 100.0 2025 2026 20.0 40.0 60.0 80.0 100.0 2025 2026

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

More in Other

All 41 in Other
  • Personal Superintelligence Index (Meta) In use Meta's composite of seven personal and knowledge-work evals: Agents' Last Exam, AutomationBench-AA, FocusBench, JobBench, GDP.pdf, VBT and DRACO.
  • Artificial Analysis Intelligence Index v5 In use Version 5 of Artificial Analysis's index: nine evaluations in four equal categories, most with private test sets, run three times each in its own agent harness. Gemini 4 Carbon 59.0
  • Anthropic ECI (AECI) Only in system cards Index that combines a basket of capability benchmarks into one capability score to track trends across Anthropic models. Claude Opus 5.5 169.4
  • Qwen in-house English automatic evaluation Only in system cards Models answer Qwen in-house English datasets on knowledge, comprehension, coding, math and related skills. Skills and weights change between reports. Qwen2-72B 72.9%
  • Recognition of model-written user turns (Anthropic) Only in system cards Models write synthetic user turns for real Claude Code transcripts, then judge which version is real. Score is accuracy above chance. Claude Mythos Preview 0.35
  • LiveBench Dropped Models answer monthly refreshed questions across math, coding, reasoning, language, instruction following, and data analysis. Scored automatically against objective ground truth. Claude Fable 5.1 85.7%