Reward Modeling on VLRewardBench (test)

84.7GeneralIXC-2.5-Reward

Evaluation Results

Method
IXC-2.5-Reward2025.01	84.7	62.5	62.9	65.8	70
UnifiedReward-7B2025.12	60.6	78.4	60.5	66.1	—
Qwen3-VL-8B Instruct + DeltaRubric2026.05	59.7	88.3	72.6	80.1	73.5
Qwen3-VL-8B Instruct + No rubric2026.05	55.8	86.1	48.3	72	63.4
ARM-Thinker-7B2025.12	55.7	72	64.8	67.8	—
Qwen3-VL-4B Instruct + DeltaRubric2026.05	55.3	87.7	65.9	77.5	69.6
LLaVA-Critic-8B2025.01	54.6	38.3	59.1	41.2	44
LLaVA-Critic-8B2026.05	54.6	38.3	59.1	44.5	45.2
Qwen3-VL-4B Instruct + No rubric2026.05	51.9	87.1	50.8	73.2	63.3
Gemini-1.5-Pro (2024-09-24)2025.01	50.8	72.5	64.2	67.2	62.5
GPT-4o (2024-08-06)2025.01	49.1	67.6	70.5	65.8	62.4
GPT-4o2025.12	49.1	67.6	70.5	65.8	—
InternVL3-8B2025.12	48.1	51.3	48.1	50	—
Gemini-1.5-Flash (2024-09-24)2025.01	47.8	59.6	58.4	57.6	55.3
InternVL3.5-8B2025.12	47.5	51.3	51.9	50.9	—
Qwen3-VL-8B2025.12	47	71.3	64.5	66	—
Qwen3-VL-8B Instruct2026.05	47	72.4	43.2	61.3	54.2
Qwen3-VL-4B Instruct2026.05	46.4	64.9	36	54.9	49.1
MM-RLHF-Reward-7B2026.05	45	50.5	57.6	50.2	51
Claude-3.5-Sonnet (2024-06-22)2025.01	43.4	55	62.3	55.3	53.6
Llama-3.2-90B2025.01	42.6	57.3	61.7	56.2	53.9
Llama-3.2-90B2026.05	42.6	57.3	61.7	56.2	53.9
GPT-4o-mini (2024-07-18)2025.01	41.7	34.5	58.2	41.5	44.8
NVLM-D-72B2025.01	38.9	31.6	62	40.1	44.1
NVLM-D-72B2026.05	38.9	31.6	62	40.1	44.1
Qwen2-VL-72B2025.01	38.1	32.8	58	39.5	43
Qwen2.5-VL-7B2025.12	37.7	48.7	60.4	50.1	—
InternVL2-8B2025.01	35.6	41.1	59	44.5	45.2
Pixtral-12B2025.01	35.6	25.9	59.9	35.8	40.4
InternVL2-8B2026.05	35.6	41.1	59	44.5	45.2
Molmo-72B2025.01	33.9	42.3	54.9	44.1	43.7
Llama-3.2-11B2025.01	33.3	38.4	56.6	42.9	42.8
Llama-3.2-11B2026.05	33.3	38.4	56.6	42.9	42.8
LLaVA-OneVision-7B-ov2025.01	32.2	20.1	57.1	29.6	36.5
Qwen2-VL-7B2025.01	31.6	19.1	51.1	28.3	33.9
Molmo-7B2025.01	31.1	31.8	56.2	37.5	39.7
Molmo-7B2026.05	31.1	31.8	56.2	37.5	39.7
VITA-1.5-7B2026.05	18.6	8.9	22.1	16.5	16.5
SliME-7B2026.05	7.2	27.1	18.6	19	17.6