Benchmark · ReasoningSettled

ARC-AGI

ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.

Updated 2026-11-17
Models tested
91
Top score
98.5
Claude Fable 5
Median
80.0
min 0.1
Top-5 spread
σ 0.0
Settled

Best score over time · one chart, every benchmark

ARC-AGI80 MODELS · FRONTIER RUNNING MAX0255075100SCORE ↑Dec 24Jun 25Dec 25May 26Nov 26RELEASE DATE →benchgecko.ai/benchmark/arc-agi · frontier
Frontier on ARC-AGI rose from 30.7 to 98.5 in 18 months · +67.8 points · latest leader Claude Fable 5 from Anthropic.
Pink dots = frontier records · 13 totalClick to open model page

91 models tested · sorted by score

#ModelScore
1Anthropic logoClaude Fable 598.5
2OpenAI logoGPT-6 Astra98.5
3OpenAI logoGPT-6.2 Astra98.5
4Google DeepMind logoGemini 4 Carbon98.5
5Anthropic logoClaude Opus 5.598.5
6OpenAI logoGPT-6.1 Sol98.5
7Google DeepMind logoGemini 3.8 Flash98.5
8Google DeepMind logoGemini 4 Argon98.5
9Anthropic logoClaude Fable 5.298.5
10Google DeepMind logoGemini 3.1 Pro Preview98.0
11Anthropic logoClaude Fable 5.197.5
12Anthropic logoClaude Opus 597.5
13OpenAI logoGPT-5.6 Sol97.5
14OpenAI logoGPT-5.6 Terra96.5
15OpenAI logoGPT-5.5 Pro96.5
16Anthropic logoClaude Sonnet 5.596.5
17Google DeepMind logoGemini 3.7 Flash95.5
18OpenAI logoGPT-6 Sol95.5
19OpenAI logoGPT-5.595.0
20OpenAI logoGPT-5.4 Pro94.5
21moonshotai logoKimi K394.5
22DeepSeek logoDeepSeek V4.1 Flash94.5
23DeepSeek logoDeepSeek V4.1 Flash 101794.5
24Anthropic logoClaude Opus 4.694.0
25OpenAI logoGPT-5.493.7
26Anthropic logoClaude Opus 4.793.5
27Anthropic logoClaude Opus 4.892.5
28Google DeepMind logoGemini 3.5 Flash92.5
29Google DeepMind logoGemini 3.6 Flash91.2
30z-ai logoGLM 5.3 Flash91.0
31OpenAI logoGPT-5.2 Pro90.5
32DeepSeek logoDeepSeek V4 Pro 081390.5
33xAI logoGrok 4.790.2
34xAI logoGrok 4.2089.5
35DeepSeek logoDeepSeek V4 Flash 073189.0
36Meta logoMuse Spark Open89.0
37Anthropic logoClaude Haiku 5.589.0
38OpenAI logoGPT-5.6 Luna88.0
39Alibaba Qwen logoQwen3.8 27B87.5
40xAI logoGrok 4.687.5
41xAI logoGrok 4.587.2
42OpenAI logoGPT-6 Luna86.7
43Anthropic logoClaude Sonnet 4.686.5
44OpenAI logoGPT-5.286.2
45Google DeepMind logoGemini 3 Flash Preview84.7
46Anthropic logoClaude Opus 4.580.0
47z-ai logoGLM 5.277.0
48Google DeepMind logoGemini 3 Pro75.0
49OpenAI logoGPT-5.172.8
50OpenAI logoGPT-5 Pro70.2
51xAI logoGrok 466.7
52OpenAI logoGPT-565.7
53moonshotai logoKimi K2.565.3
54Anthropic logoClaude Sonnet 4.563.7
55minimax logoMiniMax M2.563.7
56OpenAI logoGPT-5.4 Mini63.7
57OpenAI logoo360.8
58OpenAI logoo3 Pro59.3
59OpenAI logoo4 Mini58.7
60OpenAI logoo4 Mini (batch)58.7
61DeepSeek logoDeepSeek V3.257.0
62OpenAI logoGPT-5 Mini54.3
63Google DeepMind logoGemini 3.5 Flash Lite53.5
64OpenAI logoGPT-5.4 Nano51.5
65xAI logoGrok 4 Fast48.5
66Anthropic logoClaude Haiku 4.547.7
67z-ai logoGLM 544.7
68Google DeepMind logoGemini 2.5 Pro41.0
69Anthropic logoClaude Sonnet 440.0
70Anthropic logoClaude Opus 435.7
71OpenAI logoo3 Mini (batch)34.5
72OpenAI logoo3 Mini34.5
73Google DeepMind logoGemini 2.5 Flash32.3
74OpenAI logoo130.7
75Anthropic logoClaude 3.7 Sonnet28.6
76DeepSeek logoR1 052821.2
77OpenAI logoGPT-5 Nano20.7
78OpenAI logoo1-preview18.0
79xAI logoGrok-3 mini16.5
80DeepSeek logoR115.8
81OpenAI logoo1-mini14.0
82Alibaba Qwen logoQwen3 235B A22B Instruct 250711.0
83OpenAI logoGPT-4.510.3
84xAI logoGrok 35.5
85OpenAI logoGPT-4.15.5
86Magistral Small 1.15.0
87OpenAI logoGPT-4o (2024-11-20)4.5
88Meta logoLlama 4 Maverick4.4
89OpenAI logoGPT-4.1 Mini3.5
90Meta logoLlama 4 Scout0.5
91OpenAI logoGPT-4.1 Nano0.1

What it measures · how it is scored · sources · as of 2026-11-22

  • What it measures · Abstraction and Reasoning Corpus. Tests fluid intelligence through novel visual pattern recognition puzzles. Core measure of general intelligence.
  • How it is scored · Scored as a percentage · higher is better · maximum 100
  • Released · 2019-11-05
  • Scores from · Epoch AI Benchmarking Hub (91 scores)

Same category · related evaluations