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

Multimodal Reward Modeling on Multimodal RewardBench

88.79Accuracy

Gemini 3.1 Pro

Updated 3d ago

Evaluation Results

Method	Links
Gemini 3.1 Pro 2026.04		88.79
Gemini 3.0 Flash 2026.04		87.9
Gemini-2.5-Pro 2026.02		85.4
EGT 2026.02		84.3
Qwen3-VL-32B-Instruct 2026.04		83.02
Gemini-2.5-Flash 2026.02		82.5
Gemini 3.1 Flash-Lite 2026.04		82.25
Claude 4.6 Opus 2026.04		82.23
R1-Reward 2026.02		82.2
Qwen3-VL-30B-A3B-Instruct 2026.04		82.06
GPT-5.4 2026.04		81.91
Qwen3-VL-32B-Instruct 2026.04		81.19
Qwen3-VL-32B-Instruct 2026.04		79.35
Qwen3-VL-30B-A3B-Instruct 2026.04		75.82
Skywork-VL Reward 2026.04		74.25
Qwen3-VL-30B-A3B-Instruct 2026.04		72.49
Claude-3.7-Sonnet (2025-02-24) 2026.02		71.9
GPT-4o 2026.02		71.5
Qwen2-VL-72B 2026.02		70.9
GPT-4o (2024-08-06) 2026.02		70.8
IXC-2.5-Reward 2026.04		69.12
MM-RLHF-Reward 2026.02		67.1
IXC-2.5-Reward 2026.02		66.6
PhyCritic-7B 2026.02		65.9
Eagle-2.5-8B 2026.02		64.4
Qwen2.5-VL-7B 2026.02		64
Cosmos-R1-7B 2026.02		54.8
VITA-1.5 2026.02		53.6
Robobrain2.0-7B 2026.02		50.5
SliME 2026.02		42