Role-playing Reward Modeling on RoleRM-Bench

88.32Average ScoreRoleRM

Evaluation results

Method
RoleRM2025.12	88.32	90.74	82.54	80.28	94	90.91	91.53	88.24
internlm/internlm2-20b-reward2025.12	70.58	70.37	68.25	67.61	76	72.73	66.1	75
allenai/Llama-3.1-Tulu-3-70B-SFT-RM-RB22025.12	70.36	66.67	71.43	70.42	70	65.15	76.27	70.59
Skywork/Skywork-Reward-V2-Qwen3-8B2025.12	70.07	64.81	69.84	67.61	66	75.76	74.58	77.94
GPT-5-mini-2025-08-072025.12	69.3	68.52	73.02	59.86	83	68.94	70.34	65.44
GPT-4o-2024-08-062025.12	69.12	66.67	66.67	66.9	71	68.18	78.81	67.65
internlm/internlm2-7b-reward2025.12	67.72	64.81	63.49	64.79	68	72.73	72.88	66.18
GPT-5-2025-08-072025.12	67.55	69.44	66.67	66.2	82	65.91	60.17	62.5
allenai/Llama-3.1-Tulu-3-8B-DPO-RM-RB22025.12	67.53	70.37	65.08	60.56	76	71.21	67.8	61.76
allenai/Llama-3.1-70B-Instruct-RM-RB22025.12	66.39	72.22	65.08	56.34	62	65.15	76.27	67.65
allenai/Llama-3.1-Tulu-3-8B-RL-RM-RB22025.12	66.34	70.37	61.9	60.56	72	72.73	69.49	60.29
Claude-3-7-sonnet-202502192025.12	65.24	68.52	62.7	65.49	75	62.88	61.02	61.76
allenai/Llama-3.1-8B-Instruct-RM-RB22025.12	65.06	59.26	61.94	59.15	70	72.73	71.19	61.16
allenai/Llama-3.1-Tulu-3-8B-SFT-RM-RB22025.12	64.89	66.67	60.32	57.75	70	66.67	66.1	64.71
Skywork-Reward-V2-Llama-3.1-8B2025.12	64.17	53.7	63.49	60.56	66	71.21	69.49	64.71
CharacterRM2025.12	61.11	59.26	65.08	56.34	72	66.67	52.54	55.88
infly/INF-ORM-Llama3.1-70B2025.12	58.51	61.11	61.9	50.7	58	56.06	64.41	57.35
Ray2333/GRM_Llama3.1_8B_rewardmodel-ft2025.12	56.5	53.7	58.73	57.75	56	56.06	59.32	52.94
Skywork-Reward-Llama-3.1-8B2025.12	53.5	48.15	50.79	50.7	58	59.09	55.93	50
Skywork-Reward-Llama-3.1-8B-v0.22025.12	51.97	42.58	50.79	45.07	60	50.06	55.93	57.35
nicolinho/QRM-Llama3.1-8B-v22025.12	47.42	44.44	58.73	40.85	46	50	43.37	48.53
NCSOFT/Llama-3-OffsetBias-RM-8B2025.12	47.17	44.44	49.21	39.44	32	50	69.49	45.59