Benchmarks / Reasoning

ARC-AGI-1

ARC-AGI-1 semi-private evaluation set: novel grid-transformation puzzles measuring fluid intelligence; percent of tasks solved, with verified cost per task.

82models
–Open Index weight
–last updated
ARC Prizepublisher

ARC-AGI-1

Higher is better · best effort per model

Chart settings
82 of 82 selected

full table below

#ModelEffortScoreCost / taskHarness
1Fable 5max98.5%$2.11–
2GPT-6 Astraxhigh98.5%$0.347–
3Opus 5.5high98.5%$0.158–
4Gemini 3.8 Flashhigh98.5%$0.206–
5GPT-6.1 Solxhigh98.5%$0.074–
6Gemini 4high98.5%$0.162–
7Gemini 3.1 Prohigh98.0%$0.522–
8GPT-5.6 Solxhigh97.5%$0.400–
9Opus 5high97.5%$0.470–
10Fable 5.1max97.5%$1.40–
11GPT-5.5 Prohigh96.5%$4.53–
12GPT-5.6 Terramax96.5%$0.550–
13Gemini 3–96.0%$7.17–
14Gemini 3.7 Flashhigh95.5%$0.120–
15GPT-6 Solmax95.5%$0.141–
16GPT-5.5xhigh95.0%$0.730–
17GPT-5.4 Proxhigh94.5%$7.21–
18Kimi K3max94.5%$0.770–
19DeepSeek V4.1 Flashmax94.5%$0.066–
20Opus 4.6high94.0%$1.39–
21GPT-5.4xhigh93.7%$0.620–
22Opus 4.7high93.5%$1.41–
23Opus 4.8max92.5%$2.33–
24Gemini 3.5 Flashhigh92.5%$0.428–
25Gemini 3.6 Flashhigh91.2%$0.336–
26GLM 5.3 Flashmax91.0%$0.043–
27GPT-5.6 Lunamax90.7%$0.065–
28GPT-5.2 Proxhigh90.5%$11.65–
29DeepSeek V4 Pro 0813low90.5%$0.178–
30Grok 4.7high90.2%$0.642–
31Grok 4.20xhigh89.5%$0.350–
32DeepSeek V4 Flash 0731max89.0%$0.021–
33Haiku 5.5max89.0%$0.051–
34Grok 4.6medium87.5%$0.301–
35Qwen 3.8 27Bxhigh87.5%$0.215–
36Grok 4.5medium87.2%$0.310–
37GPT-6 Lunamax86.7%$0.018–
38Sonnet 4.6high86.5%$1.19–
39GPT-5.2xhigh86.2%$0.960–
40Gemini 3 Flashhigh84.7%$0.174–
41Inkling Smallxhigh84.0%$0.115–
42Opus 4.5high80.0%$1.47–
43Inklingmax79.5%$0.300–
44GLM 5.2max77.0%$0.192–
45Gemini 3 Prohigh75.0%$0.493–
46GPT-5.1high72.8%$0.674–
47GPT-5 Prohigh70.2%$4.78–
48Grok 4high66.7%$1.01–
49GPT-5high65.7%$0.509–
50Kimi K2.5max65.3%$0.150–
51Sonnet 4.5high63.7%$0.519–
52MiniMax M2.5high63.7%$0.070–
53GPT-5.4 Minixhigh63.7%$0.510–
54o3high60.8%$0.500–
55o4 minihigh58.7%$0.406–
56DeepSeek V3.2high57.0%$0.080–
57GPT-5 Minihigh54.3%$0.116–
58Gemini 3.5 Flash-Litehigh53.5%$0.093–
59GPT-5.4 Nanoxhigh51.5%$0.130–
60Grok 4 Fastnone48.5%$0.031–
61Haiku 4.5xhigh47.7%$0.256–
62GLM 5high44.7%$0.170–
63Gemini 2.5 Prohigh41.0%$0.484–
64Sonnet 4high40.0%$0.366–
65Opus 4medium35.7%$1.25–
66o3 minihigh34.5%$0.399–
67Gemini 2.5 Flashmedium33.3%$0.037–
68DeepSeek R1xhigh21.2%$0.046–
69GPT-5 Nanomedium20.7%$0.012–
70Grok 3 Minilow16.5%$0.0099–
71o1 minihigh14.0%$0.135–
72Qwen 3 235B-A22Bxhigh11.0%$0.0025–
73GPT-4.5none10.3%$0.290–
74Magistral Medium–6.1%$0.099–
75GPT-4.1high5.5%$0.039–
76Grok 3none5.5%$0.093–
77Magistral Small–5.0%$0.040–
78GPT-4ohigh4.5%$0.050–
79Llama 4 Maverickmedium4.4%$0.0078–
80GPT-4.1 Minihigh3.5%$0.0078–
81Llama 4 Scoutmedium0.5%$0.0041–
82GPT-4.1 Nanohigh0.0%$0.0021–