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

Multimodal Reward Modeling on MM-RLHF-RewardBench

92.4Pairwise Accuracy

Molmo2-4B Multi-response RM

Updated 4d ago

Evaluation Results

Method	Links
Molmo2-4B Multi-response RM 2026.04		92.4	-
MM-RLHF-Reward 2026.04		85	-
Qwen3-VL-4B Multi-response RM 2026.04		84.7	-
InternVL3-78B 2026.04		81.8	-
R1-Reward 2026.04		80.6	-
Qwen3-VL-4B 2026.04		80	-
Qwen3-VL-32B 2026.04		78.8	-
Qwen2.5-VL-7B 2026.04		77.6	-
LLaVA-Critic 2026.04		77.6	-
Qwen3-VL-8B 2026.04		73.5	-
Molmo2-4B 2026.04		73.5	-
Skywork-VL-Reward 2026.04		72.4	-
GPT-5 2026.04		71.8	-
IXC-2.5-Reward 2026.04		71.2	-
Gemini-2.5-Pro 2026.04		70.6	-
Claude-Sonnet-4.5 2026.04		70	-
InternVL3-8B 2026.04		69.4	-
Molmo2-8B 2026.04		68.2	-
GPT-4o (2024-08-06) 2026.02		-	58.23
Claude-3.7-Sonnet (2025-02-24) 2026.02		-	82.35
SliME 2026.02		-	17.1
VITA-1.5 2026.02		-	20.58
Qwen2-VL-72B 2026.02		-	48.23
MM-RLHF-Reward 2026.02		-	82
IXC-2.5-Reward 2026.02		-	71.18
R1-Reward 2026.02		-	80.59
EGT 2026.02		-	85.88