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

Multimodal Understanding on MMMU (dev)

67.3Accuracy

PVM-8B (SFT + GRPO)

Updated 1mo ago

Evaluation Results

Method	Links
PVM-8B (SFT + GRPO) 2026.05		67.3
PVM-8B (SFT) 2026.05		66.7
PEARL-8B 2026.05		65.3
Qwen3-VL-8B (LoRA-SFT + GRPO) 2026.05		64.7
ICoT 2026.05		63.3
Qwen3-VL-8B (LoRA-SFT) 2026.05		63.3
PVM-4B (SFT + GRPO) 2026.05		62.7
CoMemo 2026.05		62
Euclid-8B 2026.05		62
OneThinker-8B 2026.05		62
Qwen3-VL-8B (SFT) 2026.05		60.7
Qwen3-VL-8B (SFT + GRPO) 2026.05		60.7
PVM-4B (SFT) 2026.05		60.7
MemVR 2026.05		59.3
Qwen3-VL-4B (SFT) 2026.05		58
Qwen3-VL-4B (SFT + GRPO) 2026.05		58
Qwen3-VL-8B-Instruct 2026.05		57.3
Qwen3-VL-4B-Instruct 2026.05		57.3
Qwen3-VL-4B (LoRA-SFT) 2026.05		56
Qwen3-VL-4B (LoRA-SFT + GRPO) 2026.05		56
Defender 2026.01		25.33
Defender 2026.01		23.33
Base (M_def^(0)) + Clean Data 2026.01		21.33
Base (M_def^(0)) 2026.01		20.67
Defender 2026.01		20