Share your thoughts, 1 month free Claude Pro on usSee more

Abstract Reasoning on ARC-AGI 2

60Accuracy

Human panel (average)

Updated 29d ago

Evaluation Results

Method	Links
Human panel (average) 2026.06		60
Qwen3 + DIARC 2026.06		33.06
Qwen3 + SFT only 2026.06		30.83
Grok-4 2026.06		29.4
GPT-5.4 2026.06		29.2
Kimi K2.5 2026.06		11.8
GRAM 2026.06		11.1
VARC (ensemble) 2026.06		11.1
Minitron + DIARC 2026.06		10.28
Minitron + SFT only 2026.06		9.86
VARC 2026.06		8.3
Opus 4.5 2026.06		7.8
Llama-3.2 + DIARC 2026.06		6.53
HRM 2026.06		5
Llama-3.2 + SFT only 2026.06		4.44
Llama-3.2 2026.06		0
Minitron 2026.06		0
Qwen3 2026.06		0