Benchmark Almanac

Decision Index Vision

Piccolo 27B VL (1108) 75.6 tops its leaderboard
Not in model cards

Models make decisions over images and screens. The score is the published equal-weight mean of public and private vision scores.

Separate vision suite, edition 0.4, with an image cap of 1,638,400 pixels.

Released
September 2026
Kind
Composite indices
Level
Everyday
Scored
Checked automatically
Answers
Prediction
Tasks
Public, plus a private test set
88
Entries on its own leaderboard, 0 models we track

Piccolo 27B VL (1108) tops the leaderboard with 75.6

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. 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.

88 entries. Scroll sideways for the rest.

  1. 75.6 Piccolo 27B VL (1108)
  2. 75.0 JDI-Decision-V1
  3. 73.8 WeMM-JEV-27B
  4. 73.3 RBDash-Decision v0.2
  5. 73.3 Valen-Decision 1.5 27B
  6. 72.9 H2O Lightning 31B v1.3.0
  7. 72.8 Perplexity Decider v1.3 (27B)
  8. 72.5 vLLM-SR d3-turbo v1.1 (27B)
  9. 72.4 Amos 1 Large (27B)
  10. 72.1 Sieve 31B v3
  11. 71.9 AutoTrust-JEV-27B v1.1
  12. 71.7 Perplexity Decider v1.2 (27B)
  13. 71.6 StartLux-Decision v1.1 27B
  14. 71.5 vera-large-2o
  15. 71.5 Fastino GLiDE 1.2
  16. 71.4 VisionJEV 1.0.0 (27B)
  17. 71.3 vLLM-SR d3 (27B)
  18. 71.3 Nace.AI Drex Max v1 (27B)
  19. 70.6 vLLM-SR d3-ultra (Decision 3.0, 27B)
  20. 70.0 Aleria Judge 1.1.0 (Gemma 4 31B)
  21. 70.0 Herbert v0.5
  22. 69.7 Valen-Decision 1.5 9B
  23. 69.6 JEV-27B-VL
  24. 68.9 OneJev1.5-27B
  25. 68.7 jev-vision-27b
  26. 68.3 Piccolo 9B VL
  27. 68.1 KnowLine 35B-A3B Gen1
  28. 67.5 Piccolo 4B VL
  29. 66.9 vLLM-SR d3-flash (9B)
  30. 66.8 Solomon v1.1 (27B)
  31. 66.6 Valen-Decision 1.5 4B
  32. 66.0 Perplexity Decider Flash v1.2 (4B)
  33. 65.9 LingeeJevMulti-4B-v1
  34. 65.8 span-02 (Gemma 4 26B-A4B)
  35. 65.5 Dexy-Decision-9B-V2
  36. 65.2 Sieve-Vision-27B
  37. 65.2 KnowLine 31B Gen1
  38. 65.2 Perplexity Decider Flash v1.3 (4B)
  39. 64.8 vev-4b-v2
  40. 64.2 Xor 1.2 (35B-A3B)
  41. 64.0 RSI-Jev v6.1 VL 27B (effort auto)
  42. 63.6 jev-d3-46k
  43. 63.5 H2O Lightning Bolt 74B v1.1
  44. 63.4 Qwen3.5-397B (VLM)
  45. 63.4 OneJev1.5-9B
  46. 63.2 vLLM-SR d3-mini (4B)
  47. 62.0 matilda-jev-v1.5 (27B)
  48. 61.9 VISTA-4B-A
  49. 61.7 JPT-9B
  50. 60.7 Qwen3.8-27B NVFP4 (RadixArk)
  51. 59.7 Swift 1.5 NVFP4 (27B)
  52. 59.6 Valen-Decision 1.5 2B
  53. 59.5 Cloudflare clef-omni (30B-A3B)
  54. 59.0 JPT-4B
  55. 58.8 Winnow-12B
  56. 58.6 Surogate Rune 26B-A4B v3
  57. 58.6 Sieve-Vision-4B
  58. 58.0 xor-26b-a4b-v1.1
  59. 57.9 LiquidAI d1-2B
  60. 57.4 RSI-Jev v6.1 VL 4B
  61. 57.3 RSI-Jev v5.0 VL 3B
  62. 57.3 Aplomb 1
  63. 57.1 PurpleMIST-Mini-2.0
  64. 56.5 Intern-Decision-4B
  65. 56.1 Jev-Omni (12B)
  66. 55.3 QEV-2B
  67. 54.7 vLLM-SR d3-nano (2B)
  68. 53.7 Qwen3-VL-8B (VLM)
  69. 52.9 Shisa DE-2 (repeat)
  70. 52.1 LiquidAI d1-Omni-2B
  71. 52.0 H2O Lightning E4B v1.0.1
  72. 51.9 Jev-Style 2B v4
  73. 51.1 NeoHorse-Jev-4B
  74. 49.3 Valen-Decision 1.5 0.8B
  75. 47.8 Intern-Decision-2B
  76. 47.8 djev (26B-A4B)
  77. 47.7 razorback16 diffgemma (26B-A4B)
  78. 46.3 mmastrac diffgemma (26B-A4B)
  79. 44.8 QEV-0.8B
  80. 44.6 OneJev1.5-0.8B
  81. 42.1 vLLM-SR d3-lite (0.8B)
  82. 42.1 JPT-0.8B
  83. 42.1 Winnow-E4B
  84. 40.5 Saina Helm 2 Omni (0.8B)
  85. 36.3 Intern-Decision-0.8B
  86. 31.7 vLLM-SR d3-edge (0.6B)
  87. 7.0 EmbeddingGemma 2 System-One
  88. 6.2 SigLIP 2 (contrastive)
  1. 1 Piccolo 27B VL (1108) full fine-tune 75.6
  2. 2 JDI-Decision-V1 full fine-tune 75.0
  3. 3 WeMM-JEV-27B full fine-tune 73.8
  4. 4 RBDash-Decision v0.2 full fine-tune 73.3
  5. 5 Valen-Decision 1.5 27B full fine-tune 73.3
  6. 6 H2O Lightning 31B v1.3.0 full fine-tune 72.9
  7. 7 Perplexity Decider v1.3 (27B) full fine-tune 72.8
  8. 8 vLLM-SR d3-turbo v1.1 (27B) full fine-tune 72.5
  9. 9 Amos 1 Large (27B) full fine-tune 72.4
  10. 10 Sieve 31B v3 full fine-tune 72.1
  11. 11 AutoTrust-JEV-27B v1.1 full fine-tune 71.9
  12. 12 Perplexity Decider v1.2 (27B) full fine-tune 71.7
  13. 13 StartLux-Decision v1.1 27B full fine-tune 71.6
  14. 14 vera-large-2o full fine-tune 71.5
  15. 15 Fastino GLiDE 1.2 full fine-tune 71.5
  16. 16 VisionJEV 1.0.0 (27B) full fine-tune 71.4
  17. 17 vLLM-SR d3 (27B) full fine-tune 71.3
  18. 18 Nace.AI Drex Max v1 (27B) full fine-tune 71.3
  19. 19 vLLM-SR d3-ultra (Decision 3.0, 27B) full fine-tune 70.6
  20. 20 Aleria Judge 1.1.0 (Gemma 4 31B) inference technique 70.0
  21. 21 Herbert v0.5 LoRA 70.0
  22. 22 Valen-Decision 1.5 9B full fine-tune 69.7
  23. 23 JEV-27B-VL LoRA 69.6
  24. 24 OneJev1.5-27B full fine-tune 68.9
  25. 25 jev-vision-27b LoRA 68.7
  26. 26 Piccolo 9B VL full fine-tune 68.3
  27. 27 KnowLine 35B-A3B Gen1 LoRA 68.1
  28. 28 Piccolo 4B VL full fine-tune 67.5
  29. 29 vLLM-SR d3-flash (9B) full fine-tune 66.9
  30. 30 Solomon v1.1 (27B) head / adapter 66.8
  31. 31 Valen-Decision 1.5 4B full fine-tune 66.6
  32. 32 Perplexity Decider Flash v1.2 (4B) full fine-tune 66.0
  33. 33 LingeeJevMulti-4B-v1 full fine-tune 65.9
  34. 34 span-02 (Gemma 4 26B-A4B) full fine-tune 65.8
  35. 35 Dexy-Decision-9B-V2 full fine-tune 65.5
  36. 36 Sieve-Vision-27B LoRA + head 65.2
  37. 37 KnowLine 31B Gen1 LoRA 65.2
  38. 38 Perplexity Decider Flash v1.3 (4B) full fine-tune 65.2
  39. 39 vev-4b-v2 LoRA 64.8
  40. 40 Xor 1.2 (35B-A3B) LoRA 64.2
  41. 41 RSI-Jev v6.1 VL 27B (effort auto) LoRA 64.0
  42. 42 jev-d3-46k full fine-tune 63.6
  43. 43 H2O Lightning Bolt 74B v1.1 full fine-tune 63.5
  44. 44 Qwen3.5-397B (VLM) Stock reference 63.4
  45. 45 OneJev1.5-9B full fine-tune 63.4
  46. 46 vLLM-SR d3-mini (4B) full fine-tune 63.2
  47. 47 matilda-jev-v1.5 (27B) 62.0
  48. 48 VISTA-4B-A LoRA 61.9
  49. 49 JPT-9B LoRA 61.7
  50. 50 Qwen3.8-27B NVFP4 (RadixArk) inference technique 60.7
  51. 51 Swift 1.5 NVFP4 (27B) full fine-tune 59.7
  52. 52 Valen-Decision 1.5 2B full fine-tune 59.6
  53. 53 Cloudflare clef-omni (30B-A3B) full fine-tune 59.5
  54. 54 JPT-4B LoRA 59.0
  55. 55 Winnow-12B LoRA 58.8
  56. 56 Surogate Rune 26B-A4B v3 full fine-tune 58.6
  57. 57 Sieve-Vision-4B full fine-tune 58.6
  58. 58 xor-26b-a4b-v1.1 58.0
  59. 59 LiquidAI d1-2B full fine-tune 57.9
  60. 60 RSI-Jev v6.1 VL 4B full fine-tune 57.4
  61. 61 RSI-Jev v5.0 VL 3B full fine-tune 57.3
  62. 62 Aplomb 1 LoRA + head 57.3
  63. 63 PurpleMIST-Mini-2.0 full fine-tune 57.1
  64. 64 Intern-Decision-4B full fine-tune 56.5
  65. 65 Jev-Omni (12B) LoRA + head 56.1
  66. 66 QEV-2B full fine-tune 55.3
  67. 67 vLLM-SR d3-nano (2B) full fine-tune 54.7
  68. 68 Qwen3-VL-8B (VLM) Stock reference 53.7
  69. 69 Shisa DE-2 (repeat) LoRA 52.9
  70. 70 LiquidAI d1-Omni-2B full fine-tune 52.1
  71. 71 H2O Lightning E4B v1.0.1 full fine-tune 52.0
  72. 72 Jev-Style 2B v4 full fine-tune 51.9
  73. 73 NeoHorse-Jev-4B head / adapter 51.1
  74. 74 Valen-Decision 1.5 0.8B full fine-tune 49.3
  75. 75 Intern-Decision-2B full fine-tune 47.8
  76. 76 djev (26B-A4B) inference technique 47.8
  77. 77 razorback16 diffgemma (26B-A4B) inference technique 47.7
  78. 78 mmastrac diffgemma (26B-A4B) inference technique 46.3
  79. 79 QEV-0.8B full fine-tune 44.8
  80. 80 OneJev1.5-0.8B full fine-tune 44.6
  81. 81 vLLM-SR d3-lite (0.8B) full fine-tune 42.1
  82. 82 JPT-0.8B LoRA 42.1
  83. 83 Winnow-E4B LoRA 42.1
  84. 84 Saina Helm 2 Omni (0.8B) LoRA + head 40.5
  85. 85 Intern-Decision-0.8B full fine-tune 36.3
  86. 86 vLLM-SR d3-edge (0.6B) full fine-tune 31.7
  87. 87 EmbeddingGemma 2 System-One head / adapter 7.0
  88. 88 SigLIP 2 (contrastive) Stock reference 6.2

Source: Official Decision Index Vision leaderboard

#EntryScoreSettingDateSource
1 Piccolo 27B VL (1108) 75.6 full fine-tune Official leaderboard
2 JDI-Decision-V1 75.0 full fine-tune Official leaderboard
3 WeMM-JEV-27B 73.8 full fine-tune Official leaderboard
4 RBDash-Decision v0.2 73.3 full fine-tune Official leaderboard
5 Valen-Decision 1.5 27B 73.3 full fine-tune Official leaderboard
6 H2O Lightning 31B v1.3.0 72.9 full fine-tune Official leaderboard
7 Perplexity Decider v1.3 (27B) 72.8 full fine-tune Official leaderboard
8 vLLM-SR d3-turbo v1.1 (27B) 72.5 full fine-tune Official leaderboard
9 Amos 1 Large (27B) 72.4 full fine-tune Official leaderboard
10 Sieve 31B v3 72.1 full fine-tune Official leaderboard
11 AutoTrust-JEV-27B v1.1 71.9 full fine-tune Official leaderboard
12 Perplexity Decider v1.2 (27B) 71.7 full fine-tune Official leaderboard
13 StartLux-Decision v1.1 27B 71.6 full fine-tune Official leaderboard
14 vera-large-2o 71.5 full fine-tune Official leaderboard
15 Fastino GLiDE 1.2 71.5 full fine-tune Official leaderboard
16 VisionJEV 1.0.0 (27B) 71.4 full fine-tune Official leaderboard
17 vLLM-SR d3 (27B) 71.3 full fine-tune Official leaderboard
18 Nace.AI Drex Max v1 (27B) 71.3 full fine-tune Official leaderboard
19 vLLM-SR d3-ultra (Decision 3.0, 27B) 70.6 full fine-tune Official leaderboard
20 Aleria Judge 1.1.0 (Gemma 4 31B) 70.0 inference technique Official leaderboard
21 Herbert v0.5 70.0 LoRA Official leaderboard
22 Valen-Decision 1.5 9B 69.7 full fine-tune Official leaderboard
23 JEV-27B-VL 69.6 LoRA Official leaderboard
24 OneJev1.5-27B 68.9 full fine-tune Official leaderboard
25 jev-vision-27b 68.7 LoRA Official leaderboard
26 Piccolo 9B VL 68.3 full fine-tune Official leaderboard
27 KnowLine 35B-A3B Gen1 68.1 LoRA Official leaderboard
28 Piccolo 4B VL 67.5 full fine-tune Official leaderboard
29 vLLM-SR d3-flash (9B) 66.9 full fine-tune Official leaderboard
30 Solomon v1.1 (27B) 66.8 head / adapter Official leaderboard
31 Valen-Decision 1.5 4B 66.6 full fine-tune Official leaderboard
32 Perplexity Decider Flash v1.2 (4B) 66.0 full fine-tune Official leaderboard
33 LingeeJevMulti-4B-v1 65.9 full fine-tune Official leaderboard
34 span-02 (Gemma 4 26B-A4B) 65.8 full fine-tune Official leaderboard
35 Dexy-Decision-9B-V2 65.5 full fine-tune Official leaderboard
36 Sieve-Vision-27B 65.2 LoRA + head Official leaderboard
37 KnowLine 31B Gen1 65.2 LoRA Official leaderboard
38 Perplexity Decider Flash v1.3 (4B) 65.2 full fine-tune Official leaderboard
39 vev-4b-v2 64.8 LoRA Official leaderboard
40 Xor 1.2 (35B-A3B) 64.2 LoRA Official leaderboard
41 RSI-Jev v6.1 VL 27B (effort auto) 64.0 LoRA Official leaderboard
42 jev-d3-46k 63.6 full fine-tune Official leaderboard
43 H2O Lightning Bolt 74B v1.1 63.5 full fine-tune Official leaderboard
44 Qwen3.5-397B (VLM) 63.4 Stock reference Official leaderboard
45 OneJev1.5-9B 63.4 full fine-tune Official leaderboard
46 vLLM-SR d3-mini (4B) 63.2 full fine-tune Official leaderboard
47 matilda-jev-v1.5 (27B) 62.0 Official leaderboard
48 VISTA-4B-A 61.9 LoRA Official leaderboard
49 JPT-9B 61.7 LoRA Official leaderboard
50 Qwen3.8-27B NVFP4 (RadixArk) 60.7 inference technique Official leaderboard
51 Swift 1.5 NVFP4 (27B) 59.7 full fine-tune Official leaderboard
52 Valen-Decision 1.5 2B 59.6 full fine-tune Official leaderboard
53 Cloudflare clef-omni (30B-A3B) 59.5 full fine-tune Official leaderboard
54 JPT-4B 59.0 LoRA Official leaderboard
55 Winnow-12B 58.8 LoRA Official leaderboard
56 Surogate Rune 26B-A4B v3 58.6 full fine-tune Official leaderboard
57 Sieve-Vision-4B 58.6 full fine-tune Official leaderboard
58 xor-26b-a4b-v1.1 58.0 Official leaderboard
59 LiquidAI d1-2B 57.9 full fine-tune Official leaderboard
60 RSI-Jev v6.1 VL 4B 57.4 full fine-tune Official leaderboard
61 RSI-Jev v5.0 VL 3B 57.3 full fine-tune Official leaderboard
62 Aplomb 1 57.3 LoRA + head Official leaderboard
63 PurpleMIST-Mini-2.0 57.1 full fine-tune Official leaderboard
64 Intern-Decision-4B 56.5 full fine-tune Official leaderboard
65 Jev-Omni (12B) 56.1 LoRA + head Official leaderboard
66 QEV-2B 55.3 full fine-tune Official leaderboard
67 vLLM-SR d3-nano (2B) 54.7 full fine-tune Official leaderboard
68 Qwen3-VL-8B (VLM) 53.7 Stock reference Official leaderboard
69 Shisa DE-2 (repeat) 52.9 LoRA Official leaderboard
70 LiquidAI d1-Omni-2B 52.1 full fine-tune Official leaderboard
71 H2O Lightning E4B v1.0.1 52.0 full fine-tune Official leaderboard
72 Jev-Style 2B v4 51.9 full fine-tune Official leaderboard
73 NeoHorse-Jev-4B 51.1 head / adapter Official leaderboard
74 Valen-Decision 1.5 0.8B 49.3 full fine-tune Official leaderboard
75 Intern-Decision-2B 47.8 full fine-tune Official leaderboard
76 djev (26B-A4B) 47.8 inference technique Official leaderboard
77 razorback16 diffgemma (26B-A4B) 47.7 inference technique Official leaderboard
78 mmastrac diffgemma (26B-A4B) 46.3 inference technique Official leaderboard
79 QEV-0.8B 44.8 full fine-tune Official leaderboard
80 OneJev1.5-0.8B 44.6 full fine-tune Official leaderboard
81 vLLM-SR d3-lite (0.8B) 42.1 full fine-tune Official leaderboard
82 JPT-0.8B 42.1 LoRA Official leaderboard
83 Winnow-E4B 42.1 LoRA Official leaderboard
84 Saina Helm 2 Omni (0.8B) 40.5 LoRA + head Official leaderboard
85 Intern-Decision-0.8B 36.3 full fine-tune Official leaderboard
86 vLLM-SR d3-edge (0.6B) 31.7 full fine-tune Official leaderboard
87 EmbeddingGemma 2 System-One 7.0 head / adapter Official leaderboard
88 SigLIP 2 (contrastive) 6.2 Stock reference Official leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Other

All 43 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
  • SAGE v1.1 (Vals) In use Models grade handwritten undergraduate math and computer science work against rubrics. Scored by class-balanced accuracy of each rubric check against expert grading. Claude Opus 5.5 91.3%
  • MTEB Fading Massive Text Embedding Benchmark: retrieval, clustering, classification across languages and modalities. sonar 42%
  • 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
  • SPEED-Bench Only in system cards Measures speculative decoding across varied prompts and domains. Reports the average acceptance length of draft tokens, so higher means faster generation at the same output. Nemotron 3 Ultra 5.15