Skip to content

ARC-AGI-1 Semi-Private Evaluation

ARC Prize fluid-intelligence benchmark using novel visual grid transformations.

36 models

Higher is better.

  1. 1Gemini 4 ArgonGoogle98.50
  2. 2Claude Fable 5Anthropic98.50
  3. 3Gemini 3.8 FlashGoogle98.50
  4. 4Claude Opus 5.5Anthropic97.50
  5. 5GPT-6 AstraOpenAI97.50
  6. 6Claude Fable 5.1Anthropic97.50
  7. 7Claude Opus 5Anthropic97.50
  8. 8GPT-6.1 SolOpenAI96.50
  9. 9GPT-5.6 SolOpenAI96.50
  10. 10GPT-5.6 TerraOpenAI96.50
  11. 11GPT-6 SolOpenAI95.50
  12. 12Gemini 3.7 FlashGoogle95.50
  13. 13GPT-5.5 ProOpenAI95.00
  14. 14GPT-5.5OpenAI95.00
  15. 15Kimi K3Moonshot AI94.50
  16. 16GPT-5.4 ProOpenAI94.50
  17. 17GPT-5.4OpenAI93.67
  18. 18Claude Opus 4.8Anthropic92.50
  19. 19Gemini 3.5 FlashGoogle92.50
  20. 20Gemini 3.6 FlashGoogle91.17
  21. 21DeepSeek V4 Pro 0813DeepSeek90.00
  22. 22Grok 4.7xAI89.50
  23. 23Claude Haiku 5.5Anthropic89.00
  24. 24DeepSeek V4 Flash 0731DeepSeek89.00
  25. 25GPT-5.6 LunaOpenAI88.00
  26. 26Grok 4.6xAI87.00
  27. 27GPT-6 LunaOpenAI86.67
  28. 28GPT-5.2OpenAI86.17
  29. 29Claude Sonnet 4.6Anthropic86.00
  30. 30Grok 4.5xAI85.67
  31. 31Inkling-SmallThinking Machines Lab84.00
  32. 32Claude Opus 4.5 ThinkingAnthropic80.00
  33. 33Gemini 3 ProGoogle75.00
  34. 34GPT-5.1OpenAI72.83
  35. 35GPT-5.4 miniOpenAI63.70
  36. 36GPT-5.4 nanoOpenAI51.50
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Verified accuracy
Tasks
Semi-private ARC-AGI-1 evaluation set
Authors
ARC Prize Foundation (2026)

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.