List your company
ResearchFormerly WizWand

Multimodal Understanding on MMBench-CN lite

91.2AccuracyPVM-8B (SFT + GRPO)
85.79287.19688.690.004May 1, 2026
Updated 6mo ago

Evaluation results

MethodLinks
PVM-8B (SFT + GRPO)Backbone=8B, Training Strategy=SFT + GRPO, Model Architecture=Persistent Visual Memory2026.0591.2
Euclid-8BBackbone=8B, Training Strategy=RL-tuned2026.0590.4
PVM-8B (SFT)Backbone=8B, Training Strategy=SFT, Model Architecture=Persistent Visual Memory2026.0590.4
PVM-4B (SFT + GRPO)Backbone=4B, Training Strategy=SFT + GRPO, Model Architecture=Persistent Visual Memory2026.0590.4
Qwen3-VL-8B (LoRA-SFT)Backbone=8B, Training Strategy=LoRA-SFT2026.0588.8
Qwen3-VL-8B (SFT + GRPO)Backbone=8B, Training Strategy=SFT + GRPO2026.0588.8
Qwen3-VL-4B (LoRA-SFT + GRPO)Backbone=4B, Training Strategy=LoRA-SFT + GRPO2026.0588.8
ICoTTraining Strategy=Visual Injection2026.0588.2
CoMemoTraining Strategy=Visual Injection2026.0588
Qwen3-VL-8B (SFT)Backbone=8B, Training Strategy=SFT2026.0588
PVM-4B (SFT)Backbone=4B, Training Strategy=SFT, Model Architecture=Persistent Visual Memory2026.0588
Qwen3-VL-4B (SFT + GRPO)Backbone=4B, Training Strategy=SFT + GRPO2026.0588
PEARL-8BBackbone=8B, Training Strategy=RL-tuned2026.0587.2
OneThinker-8BBackbone=8B, Training Strategy=RL-tuned2026.0587.2
Qwen3-VL-4B (SFT)Backbone=4B, Training Strategy=SFT2026.0587.2
Qwen3-VL-4B (LoRA-SFT)Backbone=4B, Training Strategy=LoRA-SFT2026.0587.2
MemVRTraining Strategy=Visual Injection2026.0586.4
Qwen3-VL-8B-InstructBackbone=8B2026.0586.4
Qwen3-VL-8B (LoRA-SFT + GRPO)Backbone=8B, Training Strategy=LoRA-SFT + GRPO2026.0586.4
Qwen3-VL-4B-InstructBackbone=4B2026.0586