Inference
ARC Prize Foundation

ARC-AGI-2 (public eval)

Public set of abstract reasoning puzzles.

ModelsModels with a result.104
Top resultBest result on this test.98.3%Claude Fable 5.1
Top-3 spreadPoints from first to third.1.3 pts
YearYear of release.N/A

Result and cost

020406080100$0.003$0.01$0.03$0.1$0.3$1$3$10$30RESULTUS DOLLARS PER TASK · LOG SCALE

Results

104 results
#ModelResultThe score from the source.CostUS dollars to run one task.
01Claude Fable 5.1max effort98.3%$4.33
02GPT-6 Astramax effort97.5%$0.97
03Claude Opus 5max effort97.1%$2.07
04Claude Fable 5max effort96.4%$5.69
05GPT-5.6 Solmax effort93.8%$1.52
06GPT-5.4 Proxhigh effort92.2%$17.6
07GPT-5.6 Terramax effort91.3%$1.18
08GPT-5.5xhigh effort90.6%$1.91
09GPT-5.5 Proxhigh effort90.4%$11.1
10Gemini 3.1 Pro Preview88.1%$0.979
11Gemini 3.7 Flashhigh effort85.3%$0.273
12GPT-5.4xhigh effort84.2%$1.57
13Claude 4.7max effort82.2%$8.21
14Claude Opus 4.6 (120K, High)79.0%$3.81
15Claude Opus 4.6 (120K, Max)74.9%$3.77
16Claude Opus 4.6 (120K, Medium)73.6%$3.02
17Gemini 3.5 Flashhigh effort72.1%$0.9
18Grok 4.6xhigh effort68.3%$0.755
19Gemini 3.6 Flashhigh effort65.3%$0.658
20GPT-5.6 Luna 2026-07-30max effort64.2%$0.19
21Grok 4.20 (Reasoning)63.6%$0.96
22Claude Sonnet 4.6max effort62.4%$2.89
23GPT-5.6 Lunamax effort60.2%$0.69
24Claude Opus 4.6 (120K, Low)59.9%$2.4
25GPT-5.2xhigh effort59.8%$2.05
26Grok 4.5high effort58.2%$0.81
27GPT-5.2 Prohigh effort51.7%$16.66
28Inkling-Smallxhigh effort39.3%$0.258
29Inkling37.8%$0.713
30Gemini 3 Flash Previewhigh effort34.0%$0.24
31Opus 4.5 (Thinking, 32K)28.1%$1.37
32Opus 4.5 (Thinking, 16K)24.2%$0.82
33Grok 4 (Thinking)21.1%$2.08
34GLM-5.220.8%$0.229
35GPT-5.1 (Thinking, High)18.3%$1.22
36GPT-5.4 minixhigh effort17.8%$0.81
37Claude Sonnet 4.5 (Thinking 32K)14.7%$0.84
38GPT-5 Pro13.3%$8.01
39Kimi K2.512.1%$0.3
40Opus 4.5 (Thinking, 8K)10.4%$0.49
41GPT-5high effort9.6%$0.774
42GPT-5.1 (Thinking, Medium)8.4%$0.46
43Gemini 3.5 Flash-Litehigh effort7.9%$0.147
44o4-mini (high)high effort7.5%$0.882
45Opus 4.5 (Thinking, None)7.1%$0.22
46Claude Sonnet 4.5 (Thinking 16K)6.5%$0.356
47GPT-5 minihigh effort5.8%$0.215
48GLM-55.4%$0.29
48MiniMax M2.55.4%$0.19
50Claude Haiku 4.5 (Thinking 32K)5.1%$0.393
50Gemini 2.5 Pro (Thinking 16K)5.1%$0.722
52Tiny Recursion Model (TRM)5.0%$1.8
53Claude Sonnet 4.5 (Thinking 8K)4.9%$0.241
54Grok 4 Fast (Reasoning)4.6%$0.061
54Gemini 2.5 Pro (Thinking 32K)4.6%$0.763
56Claude Sonnet 4 (Thinking 16K)4.4%$0.51
57GPT-5.4 nanoxhigh effort4.2%$0.21
57Claude Haiku 4.5 (Thinking 16K)4.2%$0.143
59DeepSeek V3.23.9%$0.13
59o3-prohigh effort3.9%$9.15
61Claude Sonnet 4.53.8%$0.131
62Claude Opus 4 (Thinking 16K)3.3%$2.02
63o3high effort2.9%$0.9
63Claude Sonnet 4 (Thinking 8K)2.9%$0.272
65Gemini 2.5 Flash (Preview) (Thinking 24K)2.6%$0.331
65Gemini 2.5 Pro (Thinking 8K)2.6%$0.449
67Claude Opus 4 (Thinking 8K)2.5%$1.2
68GPT-5.1 (Thinking, Low)2.2%$0.135
69Claude Sonnet 42.1%$0.131
69Claude Haiku 4.5 (Thinking 8K)2.1%$0.095
69Claude Sonnet 4.5 (Thinking 1K)2.1%$0.142
72Gemini 2.5 Flash (Preview)1.9%$0.059
73Claude Opus 41.3%$0.663
73Claude Sonnet 4 (Thinking 1K)1.3%$0.147
75Gemini 2.5 Flash (Preview) (Thinking 16K)1.1%$0.324
75Gemini 2.5 Flash (Preview) (Thinking 8K)1.1%$0.205
77Qwen3-235b-a22b Instruct (25/07)0.8%<$0.01
78Claude 3.716k effort0.8%$0.57
79Claude Opus 4 (Thinking 1K)0.4%$0.758
80Gemini 2.5 Pro (Thinking 1K)0.4%$0.088
81GPT-5 nanohigh effort0.3%$0.032
81DeepSeek-R10.3%$0.08
83Deepseek R1 (05/28)0.3%$0.05
84Claude Haiku 4.50.0%$0.043
84o3-minilow effort0.0%$0.06
84Grok 3 [Beta]0.0%$0.14
84Magistral Medium0.0%$0.106
84GPT-4.10.0%$0.069
84Magistral Small0.0%$0.05
84GPT-4.1 mini0.0%$0.014
84Llama 4 Maverick0.0%$0.013
84GPT-4o0.0%$0.08
84Llama 4 Scout0.0%<$0.01
84GPT-4.1 nano0.0%<$0.01
84GPT-4o mini0.0%$0.01
84Claude Haiku 4.5 (Thinking 1K)0.0%$0.048
84Codex Mini (Latest)0.0%$0.227
84Gemini 1.5 Pro0.0%$0.04
84Gemini 2.5 Flash (Preview) (Thinking 1K)0.0%$0.035
84GPT-4.50.0%$2.07
84GPT-5.1 (Thinking, None)0.0%$0.059
84Grok 3 Minilow effort0.0%$0.013
84Magistral Medium (Thinking)0.0%$0.124
84O1 Mini0.0%$0.184