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

Visual Reasoning on HalluBench

71.85Accuracy

SaEI

Updated 1mo ago

Evaluation Results

Method	Links
SaEI 2025.12		71.85	-
NoisyRollout 2025.12		71.33	-
Shuffle-R1-Qwen-7B 2025.08		71	-
KL-Cov 2025.12		70.67	-
Vanilla GRPO 2025.12		70.48	-
ThinkLite-VL-7B 2025.08		70.2	-
NoisyRollout-7B-K12 2025.08		70.1	-
VL-Rethinker-7B 2025.08		69.9	-
MMR1-Math-7B 2025.08		69.6	-
Shuffle-R1-Qwen-3B 2025.08		69.2	-
VLAA-Thinker-7B 2025.08		67.5	-
MM-Eureka-Qwen-7B 2025.08		66.7	-
R1-OneVision-7B 2025.08		66.4	-
Qwen2.5-VL-7B 2025.08		65.2	-
Qwen2.5-VL-7B-Instruct 2025.12		64	-
R1-VL-7B 2025.08		60.9	-
Qwen2.5-VL-3B 2025.08		59.8	-
OpenVLThinker-7B 2025.08		59.1	-
Vision-R1-7B 2025.08		57.8	-
Claude-3.7-Sonnet 2025.08		55.4	-
GPT-4o 2025.08		55	-
InternVL-2.5-8B 2025.08		50.1	-
InternVL-3-8B 2025.08		49.9	-
Gemini-2.0 pro 2025.08		49.8	-
Qwen2.5-VL-7B-Instruct 2025.12		-	64
Vanilla GRPO 2025.12		-	70.18
NoisyRollout 2025.12		-	70.11
KL-Cov 2025.12		-	68.31
SaEI 2025.12		-	70.38