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

Generative tasks multi-preference setting

63Macro Acc (IA)

GPT-4.1 + RP-Reasoner

Updated 1mo ago

Evaluation Results

Method	Links
GPT-4.1 + RP-Reasoner 2026.01		63	1	26	59	59	59	1.43	2.63	2.15
GPT-5 + RP-Reasoner 2026.01		55	4	25	59	60	60	1.88	2.27	2.11
Qwen2.5-7B + RP-Reasoner 2026.01		42	3	19	45	51	49	2.13	2.88	2.58
DeepSeek-V3 + RP-Reasoner 2026.01		42	4	19	42	58	53	2.23	2.67	2.49
Qwen2.5-7B 2026.01		18	1	8	22	57	45	3.05	2.69	2.83
Qwen2.5-7B + Reminder 2026.01		8	2	5	10	52	38	3.52	2.72	3.04
Qwen2.5-7B + CoT 2026.01		0	0	0	0	53	35	4.25	3.64	3.89
DeepSeek-V3 2026.01		0	2	1	0	57	38	4.39	2.87	3.48
DeepSeek-V3 + Reminder 2026.01		0	2	1	0	58	39	4.47	2.79	3.46
DeepSeek-V3 + CoT 2026.01		0	3	2	0	57	38	4.12	3.09	3.52
GPT-4.1 2026.01		0	1	1	2	59	40	4.13	2.6	3.21
GPT-4.1 + Reminder 2026.01		0	1	1	2	56	38	4.05	2.84	3.33
GPT-4.1 + CoT 2026.01		0	1	1	0	62	42	4.23	3.19	3.61
GPT-5 2026.01		0	1	1	3	59	40	4.1	2.76	3.3
GPT-5 + Reminder 2026.01		0	2	1	2	60	39	4.03	2.63	3.23
GPT-5 + CoT 2026.01		0	1	1	0	60	37	4.26	3.31	3.73