Inference
ARC Prize Foundation

ARC-AGI-1 (public eval)

Public set of abstract reasoning puzzles.

ModelsModels with a result.98
Top resultBest result on this test.99.0%Claude Fable 5.1
Top-3 spreadPoints from first to third.0.0 pts
YearYear of release.N/A

Result and cost

020406080100$0.001$0.003$0.01$0.03$0.1$0.3$1$3$10RESULTUS DOLLARS PER TASK · LOG SCALE

Results

98 results
#ModelResultThe score from the source.CostUS dollars to run one task.
01Claude Fable 5.1max effort99.0%$0.893
01Claude Opus 5high effort99.0%$0.27
01GPT-5.6 Solmax effort99.0%$0.37
04GPT-5.4 Proxhigh effort98.3%$5.38
04GPT-5.6 Terramax effort98.3%$0.32
06GPT-5.5 Proxhigh effort98.0%$3.27
07GPT-6 Astramax effort97.8%$0.33
08Claude Fable 5max effort97.6%$1.41
09GPT-5.2 Proxhigh effort97.6%$7.72
10GPT-5.5xhigh effort97.5%$0.52
11Gemini 3.1 Pro Preview97.2%$0.373
12Claude 4.7max effort97.0%$1.77
13Claude Opus 4.6 (120K, Max)96.8%$1.3
14Gemini 3.7 Flashhigh effort96.6%$0.091
15GPT-5.4xhigh effort96.4%$0.43
16Grok 4.6xhigh effort96.3%$0.24
16Claude Opus 4.6 (120K, High)96.3%$0.873
18Gemini 3.5 Flashhigh effort96.0%$0.32
18Gemini 3.6 Flashhigh effort96.0%$0.254
20Claude Sonnet 4.6max effort95.8%$1.09
21Grok 4.20 (Reasoning)95.5%$0.24
22GPT-5.2xhigh effort95.0%$0.65
23Grok 4.5high effort94.8%$0.24
23Claude Opus 4.6 (120K, Medium)94.8%$0.618
25GPT-5.6 Luna 2026-07-30max effort91.3%$0.042
26GPT-5.6 Lunamax effort90.8%$0.18
27Inkling90.6%$0.216
28Claude Opus 4.6 (120K, Low)89.6%$0.411
29Inkling-Smallxhigh effort89.4%$0.084
30Gemini 3 Flash Previewhigh effort88.3%$0.135
31Opus 4.5 (Thinking, 32K)86.6%$0.76
32Opus 4.5 (Thinking, 16K)81.6%$0.52
33GLM-5.280.4%$0.175
34GPT-5.1 (Thinking, High)77.1%$0.546
35GPT-5 Pro77.0%$4.03
36GPT-5.4 minixhigh effort75.1%$0.35
37Claude Sonnet 4.5 (Thinking 32K)73.8%$0.442
38Grok 4 (Thinking)73.7%$0.818
39Kimi K2.573.1%$0.14
40Opus 4.5 (Thinking, 8K)70.1%$0.34
41GPT-5.1 (Thinking, Medium)68.9%$0.256
42o4-mini (high)high effort68.0%$0.325
43Gemini 3.5 Flash-Litehigh effort66.4%$0.077
44GPT-5high effort65.9%$0.392
45o3high effort64.3%$0.404
46Claude Sonnet 4.5 (Thinking 16K)63.6%$0.242
47o3-prohigh effort63.3%$3.92
48Claude Haiku 4.5 (Thinking 32K)62.9%$0.214
49DeepSeek V3.261.6%$0.07
50GPT-5 minihigh effort61.5%$0.094
51MiniMax M2.559.1%$0.09
52GLM-558.6%$0.14
53Grok 4 Fast (Reasoning)58.1%$0.028
54Claude Sonnet 4 (Thinking 16K)56.8%$0.322
55Gemini 2.5 Pro (Thinking 16K)56.4%$0.396
56Gemini 2.5 Pro (Thinking 32K)55.9%$0.416
57Claude Opus 4 (Thinking 16K)54.3%$1.14
58Claude Sonnet 4.5 (Thinking 8K)53.5%$0.168
59Opus 4.5 (Thinking, None)52.6%$0.11
60Claude Haiku 4.5 (Thinking 16K)51.4%$0.098
61Claude Sonnet 4 (Thinking 8K)48.6%$0.194
62GPT-5.4 nanoxhigh effort47.9%$0.1
63o3-minihigh effort46.6%$0.305
64Claude Opus 4 (Thinking 8K)45.6%$0.782
65Claude Haiku 4.5 (Thinking 8K)45.0%$0.064
66Gemini 2.5 Pro (Thinking 8K)44.2%$0.261
67GPT-5.1 (Thinking, Low)44.0%$0.093
67Tiny Recursion Model (TRM)44.0%$0.395
69Claude Opus 4 (Thinking 1K)43.3%$0.424
70Gemini 2.5 Flash (Preview) (Thinking 24K)39.1%$0.186
71Claude Sonnet 4.5 (Thinking 1K)36.6%$0.082
72Gemini 2.5 Flash (Preview)36.3%$0.033
73Gemini 2.5 Flash (Preview) (Thinking 16K)36.0%$0.185
74Claude Opus 435.5%$0.349
75Claude Sonnet 4.535.4%$0.069
76Gemini 2.5 Flash (Preview) (Thinking 8K)34.5%$0.116
77Codex Mini (Latest)33.4%$0.127
78Claude Sonnet 433.0%$0.07
79Claude Sonnet 4 (Thinking 1K)31.3%$0.081
80GPT-5 nanohigh effort29.7%$0.025
81Claude Haiku 4.5 (Thinking 1K)27.1%$0.027
82Deepseek R1 (05/28)27.0%$0.047
83Claude Haiku 4.526.6%$0.022
84Gemini 2.5 Flash (Preview) (Thinking 1K)18.0%$0.023
85Grok 3 Minilow effort17.6%<$0.01
86Gemini 2.5 Pro (Thinking 1K)17.5%$0.048
87Qwen3-235b-a22b Instruct (25/07)17.0%<$0.01
88O1 Mini13.2%$0.124
89GPT-5.1 (Thinking, None)12.4%$0.028
90GPT-4.111.8%$0.032
91Magistral Medium8.9%$0.107
92Magistral Small8.6%$0.029
93Grok 3 [Beta]8.4%$0.074
94Magistral Medium (Thinking)8.0%$0.106
95GPT-4.1 mini7.2%<$0.01
96Llama 4 Maverick7.1%<$0.01
97Llama 4 Scout2.4%<$0.01
98GPT-4.1 nano1.8%<$0.01