Reasoning on ARC-1 sampled 15 reasoning task instances Chollet 2019 benchmark leaderboard · SOTA2 Research

Reasoning on ARC-1 sampled 15 reasoning task instances Chollet 2019

86.47Task 1 AccuracyGPT-5

Evaluation Results

Method	Links
GPT-52025.12		86.47	—	65.05	49.61	100	89.34	100	40.27	100	96.47	54.55	82	94.72	43.76	32.6	71.95
Ours (ERO+Qwen2.5-7B)2025.12		77.65	—	78.2	98.45	78.28	90.16	100	81	100	94.61	81.82	100	96.27	71.93	73.15	70.73
Qwen2.5-7B2025.12		70.59	—	11.32	93.8	42.53	26.23	93.44	35.84	64	84.91	33.33	64	93.79	64.86	53.7	60.98
GPT-4o2025.12		61.95	—	26.99	97.67	44.34	22.95	75.41	63.8	100	88.8	60.61	64	96.89	68.61	72.05	63.41
Qwen2.5-32B2025.12		61.18	—	38.31	95.35	41.43	51.64	63.93	9.24	100	43.15	48.48	52	47.52	72.35	62.05	42.68
GPT-4o-mini2025.12		14.01	—	12	93.8	38.75	31.15	93.44	31.22	64	92.12	45.45	64	93.79	54.89	55.07	63.41