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

Visibility Reasoning on CrossPoint-Bench

96.15Object Accuracy

Human

Updated 1mo ago

Evaluation Results

Method	Links
Human 2025.12		96.15	89.66	92.73
CroPond-7B 2025.12		81.73	76.72	79.09
o3 2025.12		79.81	71.55	75.45
CroPond-3B 2025.12		79.81	65.52	72.27
Gemini-2.5-Pro 2025.12		68.27	66.38	67.27
Claude-Sonnet-4 2025.12		66.35	50.86	58.18
Qwen3-VL-30B-A3B-Instruct 2025.12		66.35	52.59	59.09
Qwen3-VL-235B-A22B-Thinking 2025.12		65.38	59.48	62.27
Qwen3-VL-235B-A22B-Instruct 2025.12		59.62	56.9	58.18
Qwen2.5-VL-32B-Instruct 2025.12		56.73	58.62	57.73
Qwen2.5-VL-3B-Instruct 2025.12		50.96	43.97	47.27
Qwen2.5-VL-7B-Instruct 2025.12		50	63.79	57.27
Qwen2.5-VL-72B-Instruct 2025.12		49.04	58.62	54.09
Qwen3-VL-30B-A3B-Thinking 2025.12		49.04	59.48	54.55
Molmo-7B-D 2025.12		49.04	51.72	50.45
RoboBrain2.0-7B 2025.12		46.15	55.17	50.91
RoboBrain2.0-7B 2025.12		46.15	52.59	49.55
RoboRefer-8B-SFT 2025.12		37.5	37.07	37.27
RoboPoint-v1-Vicuna-13B 2025.12		33.65	41.38	37.73