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

Multimodal Reward Modeling on MR2Bench Video

50.7Best-of-4 Accuracy

Molmo2-4B Multi-response RM

Updated 4d ago

Evaluation Results

Method	Links
Molmo2-4B Multi-response RM 2026.04		50.7
GPT-5 2026.04		50.1
Qwen3-VL-32B 2026.04		49.9
Gemini-2.5-Pro 2026.04		49.7
Claude-Sonnet-4.5 2026.04		49.1
IXC-2.5-Reward 2026.04		48.7
Qwen3-VL-4B 2026.04		47.9
Qwen3-VL-8B 2026.04		47.7
InternVL3-78B 2026.04		47.7
Qwen3-VL-4B Multi-response RM 2026.04		47.5
Skywork-VL-Reward 2026.04		46.7
R1-Reward 2026.04		44.9
Qwen2.5-VL-7B 2026.04		44.4
Molmo2-4B 2026.04		43.2
Molmo2-8B 2026.04		42.6
InternVL3-8B 2026.04		40.4
LLaVA-Critic 2026.04		40.2
MM-RLHF-Reward 2026.04		36.6