Inference
EN
ARC Prize Foundation

ARC-AGI-1 (semi-private)

Abstract reasoning puzzles held out from training data.

ModelsModels with a result.126
Top resultBest result on this benchmark.98.5%Gemini 4 Argon
Top-3 spreadPoints from first to third.0.0 pts
YearYear of release.—

Result and cost

126 of 126 results

Results

126 results
#ModelResultThe score from the source.CostUS dollars to run one task.
01Gemini 4 Argonhigh effort98.5%$0.162
01Claude Fable 5max effort98.5%$2.11
01Gemini 3.8 Flashhigh effort98.5%$0.206
04Gemini 3.1 Pro Preview98.0%$0.522
05Claude Fable 5.2max effort97.5%$0.921
05Gemini 4 Carbonhigh effort97.5%$0.407
05Claude Opus 5.5max effort97.5%$0.641
05GPT-6 Astramax effort97.5%$0.433
05Claude Fable 5.1max effort97.5%$1.40
05Claude Opus 5max effort97.5%$0.70
11GPT-6.1 Solmax effort96.5%$0.095
11Claude Sonnet 5.5max effort96.5%$0.502
11GPT-5.6 Solmax effort96.5%$0.54
11GPT-5.6 Terramax effort96.5%$0.55
11Inkling 2 Smallxhigh effort96.5%$0.082
16Gemini 3 Deep Think (2/26)96.0%$7.17
17GPT-6 Solmax effort95.5%$0.141
17Gemini 3.7 Flashhigh effort95.5%$0.12
19GPT-5.5 Proxhigh effort95.0%$4.52
19GPT-5.5xhigh effort95.0%$0.73
21GPT-5.4 Proxhigh effort94.5%$7.21
21DeepSeek V4.1 Flash 1017max effort94.5%$0.066
21Kimi K3max effort94.5%$0.77
21GPT-5.2 (Refinement)94.5%$11.40
25Claude Opus 4.6 (120K, High)94.0%$1.39
26GPT-5.4xhigh effort93.7%$0.62
27Claude Opus 4.6 (120K, Max)93.0%$1.88
28Claude Opus 4.8max effort92.5%$2.33
28Gemini 3.5 Flashhigh effort92.5%$0.428
30Claude 4.7max effort92.0%$2.58
30Claude Opus 4.6 (120K, Medium)92.0%$0.965
32Gemini 3.6 Flashhigh effort91.2%$0.336
33GLM-5.3-Flashmax effort91.0%$0.043
34GPT-5.6 Luna 2026-07-30max effort90.7%$0.065
35GPT-5.2 Proxhigh effort90.5%$11.65
36DeepSeek V4 Pro 0813max effort90.0%$0.296
37Grok 4.7xhigh effort89.5%$0.778
37Grok 4.20 (Reasoning)89.5%$0.35
39Claude Haiku 5.5max effort89.0%$0.051
39Muse Spark Openxhigh effort89.0%$0.287
39DeepSeek V4 Flash 0731max effort89.0%$0.021
42GPT-5.6 Lunamax effort88.0%$0.32
43Qwen3.8-27Bxhigh effort87.5%$0.215
43Gemini 3 Deep Think (Preview) ²87.5%$44.26
45Grok 4.6xhigh effort87.0%$0.348
46GPT-6 Lunamax effort86.7%$0.018
47GPT-5.2xhigh effort86.2%$0.96
48Claude Sonnet 4.6max effort86.0%$1.45
48Claude Opus 4.6 (120K, Low)86.0%$0.648
50Grok 4.5high effort85.7%$0.33
51Gemini 3 Flash Previewhigh effort84.7%$0.174
52Inkling-Smallxhigh effort84.0%$0.115
53Grok 4 (Refinement)79.6%$8.42
54Inkling79.5%$0.30
55GLM-5.277.0%$0.192
56Gemini 3 Pro75.0%$0.493
57Beam 1.0xhigh effort74.0%$0.306
58Claude Opus 4.5 Thinking16k effort73.9%$0.62
59GPT-5.1 (Thinking, High)72.8%$0.674
60GPT-5 Pro70.2%$4.78
61Grok 4 (Thinking)66.7%$1.01
62GPT-5high effort65.7%$0.509
63Kimi K2.565.3%$0.15
64GPT-5.4 minixhigh effort63.7%$0.51
64MiniMax M2.563.7%$0.07
64Claude Sonnet 4.5 (Thinking 32K)63.7%$0.519
67o3high effort60.8%$0.50
68o3-prohigh effort59.3%$4.16
69o4-minihigh effort58.7%$0.406
70GPT-5.1 (Thinking, Medium)57.7%$0.315
71DeepSeek V3.257.0%$0.08
72GPT-5 minihigh effort54.3%$0.116
73Gemini 3.5 Flash-Litehigh effort53.5%$0.093
74GPT-5.4 nanoxhigh effort51.5%$0.13
75Grok 4 Fast (Reasoning)48.5%$0.031
76Claude Sonnet 4.5 (Thinking 16K)48.3%$0.268
77Claude Haiku 4.5 (Thinking 32K)47.7%$0.256
78Claude Sonnet 4.5 (Thinking 8K)46.5%$0.178
79GLM-544.7%$0.17
80Gemini 2.5 Pro (Thinking 16K)41.0%$0.484
81Claude Sonnet 4 (Thinking 16K)40.0%$0.366
81Tiny Recursion Model (TRM)40.0%$1.76
83Claude Haiku 4.5 (Thinking 16K)37.3%$0.104
84Gemini 2.5 Pro (Thinking 32K)37.0%$0.512
85Claude Opus 4 (Thinking 16K)35.7%$1.25
86o3-minihigh effort34.5%$0.399
87Gemini 2.5 Flash (Preview)33.3%$0.037
87Gemini 2.5 Flash (Preview) (Thinking 16K)33.3%$0.213
89GPT-5.1 (Thinking, Low)33.2%$0.10
90Gemini 2.5 Flash (Preview) (Thinking 24K)32.3%$0.197
91Claude Sonnet 4.5 (Thinking 1K)31.0%$0.093
92Claude Opus 4 (Thinking 8K)30.7%$0.741
93Gemini 2.5 Pro (Thinking 8K)29.5%$0.295
94Claude Sonnet 4 (Thinking 8K)29.0%$0.195
95Claude Sonnet 4 (Thinking 1K)28.0%$0.094
96Codex Mini (Latest)27.3%$0.16
97Claude Opus 4 (Thinking 1K)27.0%$0.502
98Gemini 2.5 Flash (Preview) (Thinking 8K)25.8%$0.134
99Claude Sonnet 4.525.5%$0.081
99Claude Haiku 4.5 (Thinking 8K)25.5%$0.069
101Claude Sonnet 423.8%$0.081
102Nemotron 3.5 Super VLhigh effort22.7%$0.031
103Claude Opus 422.5%$0.404
104DeepSeek-R1-052821.2%$0.046
104Claude 3.78k effort21.2%$0.21
106Claude Haiku 4.5 (Thinking 1K)16.8%$0.031
107GPT-5 nanohigh effort16.7%$0.029
108Grok 3 Minilow effort16.5%$0.0099
109Gemini 2.5 Flash (Preview) (Thinking 1K)16.0%$0.036
109Gemini 2.5 Pro (Thinking 1K)16.0%$0.057
111DeepSeek-R115.8%$0.06
112Claude Haiku 4.514.3%$0.026
113o1-mini14.0%$0.135
114Qwen3 235B 250711.0%$0.0025
115GPT-4.510.3%$0.29
116Magistral Medium (Thinking)6.1%$0.099
117Magistral Medium5.9%$0.102
118GPT-5.1 Thinkingnone effort5.8%$0.034
119Grok 3 [Beta]5.5%$0.093
119GPT-4.15.5%$0.039
121Magistral Small5.0%$0.04
122GPT-4o4.5%$0.05
123Llama 4 Maverick4.4%$0.0078
124GPT-4.1 mini3.5%$0.0078
125Llama 4 Scout0.5%$0.0041
126GPT-4.1 nano0.0%$0.0021