List your company

Visual Question Answering on OCRBench v2

72.7ScoreCLAIMDIFF-RL actor-only
67.91669.15870.471.642May 19, 2026

Evaluation results

MethodLinks
CLAIMDIFF-RL actor-onlyReward=actor-only, Initialization=Qwen2026.0572.7
Qwen3VL-32B-Instruct2026.0571.7
CLAIMDIFF-RL relativeReward=relative, Initialization=SFT2026.0570.34
CLAIMDIFF-RL actor-onlyReward=actor-only, Initialization=SFT2026.0569.24
HOLISTIC-RLReference Setting=w/ ref, Initialization=SFT2026.0568.7
SFT2026.0568.38
HOLISTIC-RLReference Setting=w/o ref, Initialization=SFT2026.0568.1