Inference
EN
ARC Prize Foundation

ARC-AGI-2 (semi-private)

Abstract reasoning puzzles held out from training data.

ModelsModels with a result.131
Top resultBest result on this benchmark.95.0%GPT-6 Astra
Top-3 spreadPoints from first to third.2.5 pts
YearYear of release.—

Result and cost

131 of 131 results

Results

131 results
#ModelResultThe score from the source.CostUS dollars to run one task.
01GPT-6 Astramax effort95.0%$1.12
02GPT-6.1 Solmax effort94.2%$0.254
03Claude Fable 5.2max effort92.5%$2.81
03GPT-5.6 Solmax effort92.5%$1.44
05Gemini 4 Carbonhigh effort92.1%$0.684
06Claude Opus 5.5max effort91.7%$1.85
07Gemini 4 Argonhigh effort91.3%$0.325
08Claude Opus 5max effort90.4%$2.06
09Claude Fable 5.1max effort90.0%$4.49
10GPT-6 Solmax effort89.6%$0.439
11Claude Fable 5max effort89.2%$5.45
11Gemini 3.8 Flashhigh effort89.2%$0.40
13GPT-5.5xhigh effort85.0%$1.87
14Gemini 3.7 Flashhigh effort84.6%$0.249
14Gemini 3 Deep Think (2/26)84.6%$13.62
16GPT-5.5 Proxhigh effort84.2%$10.76
17GPT-5.6 Terramax effort83.9%$1.09
17Claude Sonnet 5.5xhigh effort83.9%$0.546
19GPT-5.4 Proxhigh effort83.3%$16.41
20Inkling 2 Smallxhigh effort81.7%$0.272
21Gemini 3.1 Pro Preview77.1%$0.962
22dots3-note Previewmax effort76.8%$0.077
23Claude 4.7max effort75.8%$7.43
24GPT-5.4xhigh effort74.0%$1.52
25DeepSeek V4.1 Flash 1017max effort72.9%$0.129
25GPT-5.2 (Refinement)72.9%$38.99
27Claude Opus 4.8high effort72.1%$2.74
27Gemini 3.5 Flashhigh effort72.1%$0.85
29Claude Opus 4.6 (120K, High)69.2%$3.47
30Claude Opus 4.6 (120K, Max)68.8%$3.64
31Grok 4.6xhigh effort67.1%$0.757
32Claude Opus 4.6 (120K, Medium)66.3%$2.72
33GLM-5.3-Flashmax effort65.8%$0.093
34Grok 4.20 (Reasoning)65.1%$0.92
35Claude Opus 4.6 (120K, Low)64.6%$2.25
36Grok 4.7xhigh effort61.4%$2.01
36DeepSeek V4 Flash 0731max effort61.4%$0.042
38Claude Haiku 5.5xhigh effort61.3%$0.085
38DeepSeek V4 Pro 0813max effort61.3%$0.598
40Kimi K3max effort60.4%$1.59
40Gemini 3.6 Flashhigh effort60.4%$0.606
42GPT-5.6 Luna 2026-07-30max effort59.6%$0.177
43GPT-5.6 Lunamax effort59.5%$0.67
44GPT-6 Lunamax effort59.3%$0.062
45Claude Sonnet 4.6max effort58.3%$2.72
46Muse Spark Openxhigh effort57.5%$0.728
47GPT-5.2 Prohigh effort54.2%$15.72
48Gemini 3 Pro (Refinement)54.0%$30.57
49GPT-5.2xhigh effort52.9%$1.90
50Grok 4.5high effort52.6%$0.78
51Gemini 3 Deep Think (Preview) ²45.1%$77.16
52Qwen3.8-27Bxhigh effort42.4%$0.447
53Inkling-Smallxhigh effort40.1%$0.233
54Inkling36.5%$0.642
55Gemini 3 Flash Previewhigh effort33.6%$0.231
56Gemini 3 Pro31.1%$0.811
57Grok 4 (Refinement)29.4%$30.40
58Beam 1.0xhigh effort23.5%$0.61
59Claude Opus 4.5 Thinking16k effort22.8%$0.79
59GLM-5.222.8%$0.25
61GPT-5.4 minixhigh effort18.9%$0.75
62GPT-5 Pro18.3%$7.14
63GPT-5.1 (Thinking, High)17.6%$1.17
64Grok 4 (Thinking)16.0%$2.17
65Claude Sonnet 4.5 (Thinking 32K)13.6%$0.759
66Kimi K2.511.8%$0.28
67Gemini 3.5 Flash-Litehigh effort10.3%$0.139
68GPT-5high effort9.9%$0.73
69Claude Opus 4 (Thinking 16K)8.6%$1.93
70Claude Sonnet 4.5 (Thinking 16K)6.9%$0.35
70Claude Sonnet 4.5 (Thinking 8K)6.9%$0.235
72o3high effort6.5%$0.834
72GPT-5.1 (Thinking, Medium)6.5%$0.421
74Tiny Recursion Model (TRM)6.3%$2.10
75o4-minihigh effort6.1%$0.856
76Claude Sonnet 4 (Thinking 16K)5.9%$0.486
77Claude Sonnet 4.5 (Thinking 1K)5.8%$0.142
78GPT-5.4 nanoxhigh effort5.7%$0.16
79Grok 4 Fast (Reasoning)5.3%$0.061
80GLM-54.9%$0.27
80MiniMax M2.54.9%$0.17
80Gemini 2.5 Pro (Thinking 32K)4.9%$0.757
80o3-prohigh effort4.9%$7.55
84Claude Opus 4 (Thinking 8K)4.5%$1.16
85GPT-5 minihigh effort4.4%$0.198
86DeepSeek V3.24.0%$0.12
86Claude Haiku 4.5 (Thinking 32K)4.0%$0.377
86Gemini 2.5 Pro (Thinking 16K)4.0%$0.715
89Claude Sonnet 4.53.8%$0.13
90o3-minihigh effort3.0%$0.547
91Gemini 2.5 Pro (Thinking 8K)2.9%$0.444
92Claude Haiku 4.5 (Thinking 16K)2.8%$0.139
93GPT-5 nanohigh effort2.6%$0.03
94Gemini 2.5 Flash (Preview) (Thinking 24K)2.5%$0.319
95Gemini 2.5 Flash (Preview) (Thinking 1K)2.2%$0.03
96Claude Sonnet 4 (Thinking 8K)2.1%$0.265
96Gemini 2.5 Flash (Preview) (Thinking 8K)2.1%$0.199
98Gemini 2.5 Flash (Preview) (Thinking 16K)2.0%$0.317
99GPT-5.1 (Thinking, Low)1.9%$0.129
100Gemini 2.5 Flash (Preview)1.7%$0.057
100Claude Haiku 4.5 (Thinking 8K)1.7%$0.091
102Nemotron 3.5 Super VLhigh effort1.5%$0.043
103DeepSeek-R11.3%$0.08
103Gemini 2.0 Flash1.3%$0.004
103Claude Opus 41.3%$0.639
103Claude Sonnet 41.3%$0.127
103Codex Mini (Latest)1.3%$0.23
103Claude Haiku 4.51.3%$0.043
103Qwen3 235B 25071.3%$0.0044
103Claude Haiku 4.5 (Thinking 1K)1.3%$0.047
111DeepSeek-R1-05281.1%$0.053
112Claude Sonnet 4 (Thinking 1K)0.9%$0.143
113o1-mini0.8%$0.191
113GPT-4.50.8%$2.10
113Gemini 1.5 Pro0.8%$0.04
116Claude 3.716k effort0.7%$0.51
117Grok 3 Minilow effort0.4%$0.013
117GPT-4.10.4%$0.069
117GPT-5.1 Thinkingnone effort0.4%$0.058
120Grok 3 [Beta]0.0%$0.142
120GPT-4.1 mini0.0%$0.014
120Llama 4 Maverick0.0%$0.012
120Magistral Medium0.0%$0.108
120Magistral Small0.0%$0.049
120GPT-4o0.0%$0.08
120Llama 4 Scout0.0%$0.0062
120GPT-4.1 nano0.0%$0.0036
120GPT-4o mini0.0%$0.01
120Claude Opus 4 (Thinking 1K)0.0%$0.75
120Gemini 2.5 Pro (Thinking 1K)0.0%$0.089
120Magistral Medium (Thinking)0.0%$0.123