evals.report
BenchmarksLabsCompareRun guidesIn the wild
BenchmarksReasoning

ARC-AGI-2

The ARC-AGI-2 abstract-reasoning puzzle benchmark (semi-private set), the harder static successor to ARC-AGI-1.

ReasoningaccuracyHigher is better
ScoresAboutRun this benchmark

What is ARC-AGI-2?

The ARC-AGI-2 abstract-reasoning puzzle benchmark (semi-private set), the harder static successor to ARC-AGI-1. evals.report tracks reported ARC-AGI-2 scores with the model, source, status, date, and run caveats attached — official leaderboard scores, vendor-reported launches, and clearly labeled community runs.

Top reported ARC-AGI-2 score: GPT-6 Astra — 95% (accuracy).

ModelLabScore↓Source modelStatusDate
GPT-6 AstraOpenAI95%GPT-6 Astra (Max)OfficialSep 3, 2026Details
Claude Opus 5.5Anthropic93.33%Claude Opus 5.5 (High)OfficialSep 22, 2026Details
GPT-5.6 SolOpenAI92.5%GPT-5.6 Sol (Max)OfficialJul 9, 2026Details
Claude Opus 5Anthropic90.42%Claude Opus 5 (Max)OfficialJul 24, 2026Details
Claude Fable 5.1Anthropic90%Claude Fable 5.1 (Max)OfficialSep 1, 2026Details
Claude Fable 5Anthropic89.17%Claude Fable 5 (Max)OfficialJun 9, 2026Details
GPT-5.5OpenAI85%GPT-5.5 (xHigh)OfficialApr 23, 2026Details
Gemini 3.7 FlashGoogle DeepMind84.58%Gemini 3.7 Flash (High)OfficialAug 13, 2026Details
Gemini 3 Deep ThinkGoogle DeepMind84.58%Gemini 3 Deep Think (2/26)OfficialDec 4, 2025Details
GPT-5.5 ProOpenAI84.58%GPT-5.5 Pro (High)OfficialApr 23, 2026Details
GPT-5.6 TerraOpenAI83.9%GPT-5.6 Terra (Max)OfficialJul 9, 2026Details
GPT-5.4 ProOpenAI83.33%GPT-5.4 Pro (xHigh)OfficialMar 5, 2026Details
Gemini 3.1 Pro PreviewGoogle DeepMind77.08%Gemini 3.1 Pro (Preview)OfficialFeb 19, 2026Details
dots3-note PreviewOpenDots Studio76.81%dots3-note Preview (Max)OfficialAug 14, 2026Details
Claude Opus 4.7Anthropic75.83%Claude 4.7 (Max)OfficialApr 16, 2026Details
GPT-5.4OpenAI73.95%GPT-5.4 (xHigh)OfficialMar 5, 2026Details
Claude Opus 4.8Anthropic72.08%Claude Opus 4.8 (High)OfficialMay 28, 2026Details
Gemini 3.5 FlashGoogle DeepMind72.08%Gemini 3.5 Flash (High)OfficialMay 19, 2026Details
Claude Opus 4.6Anthropic69.17%Claude Opus 4.6 (120K, High)OfficialFeb 5, 2026Details
Grok 4.6xAI67.08%Grok 4.6 (XHigh)OfficialAug 12, 2026Details
Grok 4.20 beta reasoningxAI65.14%Grok 4.20 (Reasoning)OfficialMar 9, 2026Details
DeepSeek V4 Flash 0731OpenDeepSeek61.39%DeepSeek V4 Flash 0731 (Max)OfficialJul 31, 2026Details
DeepSeek V4 Pro 0813OpenDeepSeek61.25%DeepSeek V4 Pro 0813 (Max)OfficialAug 13, 2026Details
Kimi K3OpenMoonshot AI60.42%Kimi K3 (Max)OfficialJul 16, 2026Details
Gemini 3.6 FlashGoogle DeepMind60.42%Gemini 3.6 Flash (High)OfficialJul 21, 2026Details
Claude Sonnet 4.6Anthropic60.42%Claude Sonnet 4.6 (High)OfficialFeb 17, 2026Details
GPT-5.6 LunaOpenAI59.54%GPT-5.6 Luna (Max)OfficialJul 9, 2026Details
GPT-6 LunaOpenAI59.31%GPT-6 Luna (Max)OfficialSep 22, 2026Details
GPT-5.2OpenAI52.91%GPT-5.2 (xHigh)OfficialDec 11, 2025Details
Grok 4.5xAI52.64%Grok 4.5 (High)OfficialJul 8, 2026Details
Inkling SmallOpenThinking Machines Lab40.14%Inkling Small (XHigh)OfficialJul 30, 2026Details
Claude Opus 4.5Anthropic37.64%Opus 4.5 (Thinking, 64K)OfficialNov 24, 2025Details
InklingOpenThinking Machines Lab36.53%InklingOfficialJul 15, 2026Details
Gemini 3 FlashGoogle DeepMind33.61%Gemini 3 Flash Preview (High)OfficialDec 17, 2025Details
Gemini 3 ProGoogle DeepMind31.11%Gemini 3 ProOfficialNov 18, 2025Details
GLM-5.2OpenZ.ai22.78%GLM-5.2OfficialJun 16, 2026Details
GPT-5.1OpenAI17.64%GPT-5.1 (Thinking, High)OfficialNov 12, 2025Details
Grok 4xAI15.97%Grok 4 (Thinking)OfficialJul 9, 2025Details
Claude Sonnet 4.5Anthropic13.61%Claude Sonnet 4.5 (Thinking 32K)OfficialSep 29, 2025Details
Kimi K2.5OpenMoonshot AI11.81%Kimi K2.5OfficialJan 27, 2026Details
Gemini 3.5 Flash-LiteGoogle DeepMind10.28%Gemini 3.5 Flash-Lite (High)OfficialJul 21, 2026Details
GPT-5OpenAI9.86%GPT-5 (High)OfficialAug 7, 2025Details
o3OpenAI6.53%o3 (High)OfficialApr 16, 2025Details
o4-miniOpenAI6.11%o4-mini (High)OfficialApr 16, 2025Details
Claude Sonnet 4Anthropic5.93%Claude Sonnet 4 (Thinking 16K)OfficialMay 22, 2025Details
GLM-5OpenZ.ai4.86%GLM-5OfficialFeb 11, 2026Details
GPT-5 miniOpenAI4.44%GPT-5 Mini (High)OfficialAug 7, 2025Details
Claude Haiku 4.5Anthropic4.03%Claude Haiku 4.5 (Thinking 32K)OfficialOct 15, 2025Details
DeepSeek V3.2OpenDeepSeek4.03%Deepseek V3.2OfficialDec 1, 2025Details

Each row reports the model’s accuracy on ARC-AGI-2. Click a row for the full run context.