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

High-resolution Visual Understanding on HR-Bench 8K

95FSP

RTWI

Updated 9d ago

Evaluation Results

Method
RTWI 2026.02	95	76	-
SC 2026.02	93	74.8	-
CISC 2026.02	93	75.8	-
Deepconf 2026.02	93	75.8	-
Self-Cer. 2026.02	92.5	75.5	-
ZwZ-Qwen2.5-VL-7B + Q-Zoom 2026.04	92.5	66.3	79.4
Zoom-Refine 2026.03	92	60	76
ZwZ-Qwen3-VL-4B + Q-Zoom 2026.04	91.3	69.8	80.5
Base 2026.02	90.8	71.3	-
Qwen2.5-VL-7B + Q-Zoom 2026.04	90.8	63.8	77.3
TikArt w/o Segment Action 2026.02	89.25	61.5	75.38
RTWI 2026.02	89	63	-
CISC 2026.02	88.5	56	-
ZoomEye 2025.05	88.5	50	69.3
Self-Cer. 2026.02	88	59	-
ZoomEye 2026.03	88	60	74
ZoomEye 2026.03	88	53	70.5
Qwen3-VL-4B + SD-RPN 2026.04	87.5	64.8	76.1
Qwen3-VL-4B + Q-Zoom 2026.04	87.5	64.8	76.1
Qwen2.5-VL-3B + Q-Zoom 2026.04	87.3	55.5	71.4
Deepconf 2026.02	87	61.8	-
Zoom-Refine 2026.03	87	63	75
DeepEyes 2025.05	86.8	58.5	72.6
GRPO 2026.03	86.75	68.25	77.5
Qwen2.5-VL-7B + DeepEyes 2026.04	86.6	58.5	72.6
SC 2026.02	86.5	61	-
Qwen2.5-VL-7B + Thyme 2026.04	86.5	57.5	72
Qwen2.5-VL-7B + SD-RPN 2026.04	86.5	60.5	73.5
TikArt w/o Observation 2026.02	85.25	64.75	75
TikArt w/o Zoom Action 2026.02	85.25	64.25	74.75
SIEVE 2026.03	85	67.25	76.13
Qwen3-VL-Instruct 2026.02	84.75	68	76.38
Gemini-2.5-pro 2026.02	84.5	82.75	83.62
TikArt 2026.02	84.5	68.25	76.38
Qwen2.5-VL* 2025.05	84.5	56.3	70.4
GRPO 2026.03	83	67.75	75.38
SIEVE 2026.03	83	73.5	78.25
Qwen3-VL-235B-A22B-Instruct 2026.02	82.25	70	76.13
Qwen3-VL-Instruct 2026.02	81.25	55.75	68.5
Vanilla* 2026.03	81	67.5	74.25
Vanilla* 2026.03	80.75	64	72.38
Pixel-Reasoner 2025.05	80	54.3	66.9
Monet-7B 2026.02	79.75	56.25	68
Qwen2.5-VL-3B + SD-RPN 2026.04	79.5	51	65.3
Gemini-2.5-flash 2026.02	79	73.5	76.25
Qwen2.5-VL* 2025.05	78.8	51.8	65.3
Base 2026.02	78.3	56	-
Deepeyes 2026.02	77	53.25	65.13
ZwZ-Qwen3-VL-4B 2026.04	77	65.8	71.4
TikArt w/o GRPO 2026.02	76.5	70.25	73.38
ZwZ-Qwen2.5-VL-7B 2026.04	73.8	57	65.4
Qwen3-VL-Thinking 2026.02	73.75	60.5	67.13
Qwen3-VL-4B 2026.04	73.5	64	68.8
Qwen2.5-VL-7B 2026.04	72.8	54.5	63.6
DyFo 2026.03	71.5	54.5	63
GPT-5 2026.02	70.5	72.75	71.62
Qwen2.5-VL-3B 2026.04	70	47.8	58.9
DyFo 2026.03	69.75	53.5	61.62
LLaVA-OneVision 2025.05	67.3	52.3	59.8
GPT-5-nano 2026.02	65.75	61.5	63.63
GPT-4o 2025.05	62	49	55.5
GPT-4o 2026.02	60.75	60	60.38
LLaVA-OneVision-1.5 2026.02	59.5	43.75	51.62
LLaVA-1.5-13B + SD-RPN 2026.04	51.8	39.8	45.8
LLaVA-1.5-7B + SD-RPN 2026.04	48.8	34.5	41.6
LLaVA-1.5-7B + Q-Zoom 2026.04	48.5	34.5	41.4
LLaVA-1.5-13B + ViCrop 2026.04	44.5	37	40.6
LLaVA-1.5-13B + S2 2026.04	41.3	44.5	42.9
LLaVA-1.5-7B + S2 2026.04	40.5	36.5	38.5
LLaVA-1.5-7B + ViCrop 2026.04	38.5	33.8	36.1
LLaVA-1.5-13B + Q-Zoom 2026.04	36.8	51	43.9
LLaVA-1.5-13B 2026.04	36.6	38.5	36.6
LLaVA-1.5-7B 2026.04	32.5	33.8	33.8
Pixel-Reasoner 2026.02	-	-	66.9
Thyme 2026.02	-	-	72
Qwen2.5-VL-7B + DeepEyesv2 2026.04	-	-	73.8