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

Generative Task Evaluation on Generative tasks single-preference setting

78Accuracy (Ignored)

Qwen2.5-7B + RP-Reasoner

Updated 1mo ago

Evaluation Results

Method	Links
Qwen2.5-7B + RP-Reasoner 2026.01		78	78	62	73	0.66	0.76	1.62	1.01
DeepSeek-V3 + RP-Reasoner 2026.01		72	94	98	88	1.06	0.32	0.28	0.55
GPT-5 + RP-Reasoner 2026.01		72	100	100	91	1.52	0.98	1.06	1.19
GPT-4.1 + RP-Reasoner 2026.01		70	98	100	89	1.1	0.96	0.9	0.99
Qwen2.5-7B 2026.01		46	94	60	67	1.94	1.08	1.96	1.66
Qwen2.5-7B + Reminder 2026.01		38	96	66	67	2.2	1.04	1.72	1.65
GPT-5 + Reminder 2026.01		8	100	100	69	3.08	1	1.04	1.71
GPT-5 2026.01		7	100	100	61	3.22	1.02	1.07	1.79
GPT-4.1 2026.01		6	100	100	69	3.28	0.9	0.98	1.72
DeepSeek-V3 + Reminder 2026.01		4	98	98	67	3.72	1.06	0.92	1.9
Qwen2.5-7B + CoT 2026.01		2	98	88	63	3.74	1.2	1.74	2.23
DeepSeek-V3 2026.01		2	100	96	66	3.98	0.98	0.96	1.97
GPT-4.1 + Reminder 2026.01		2	100	100	67	3.24	1	0.94	1.73
DeepSeek-V3 + CoT 2026.01		0	100	100	67	3.84	1.1	1.64	2.19
GPT-4.1 + CoT 2026.01		0	100	98	66	3.9	1.24	1.52	2.22
GPT-5 + CoT 2026.01		0	100	100	67	3.88	1.12	1.5	2.16