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

Visual Reasoning on HallusionBench

73.2Accuracy

GSPO + DyCo-RL

Updated 22d ago

Evaluation Results

Method	Links
GSPO + DyCo-RL 2026.06		73.2	-
GRPO + DyCo-RL 2026.06		71.6	-
GSPO 2026.06		71.4	-
DAPO + DyCo-RL 2026.06		71	-
SAPO 2026.06		71	-
DAPO 2026.06		70	-
GRPO 2026.06		69.2	-
SAPO + DyCo-RL 2026.06		69.2	-
Ours 2025.10		68.19	68.32
VCD 2025.10		65.68	67.91
GPT5 2026.07		65.2	-
Zero-Shot 2026.06		65	-
Vanilla 2025.10		64.76	64.97
Gemini-2.5-Pro 2026.07		64.1	-
SAPO + DyCo-RL 2026.06		63.9	-
DAPO + DyCo-RL 2026.06		62.7	-
GRPO + DyCo-RL 2026.06		62.4	-
GSPO 2026.06		62.4	-
GSPO + DyCo-RL 2026.06		62	-
AGLA 2025.10		61.36	63.51
Zero-Shot 2026.06		61.3	-
GRPO 2026.06		60.8	-
Ours 2025.10		60.77	60.88
DAPO 2026.06		60.7	-
LASER 2026.07		60.7	-
SAPO 2026.06		59.7	-
VCD 2025.10		58.96	59.03
Vanilla 2025.10		58.26	58.49
VAPO-Thinker-7B 2026.07		57.4	-
AGLA 2025.10		55.69	55.75
Reflection-V-7B 2026.07		53.9	-
Ours 2025.10		53.64	53.77
Qwen2.5-VL-7B 2026.07		53.6	-
R1-Onevision-7B 2026.07		52.5	-
OpenVLThinker-7B 2026.07		52.2	-
CGD 2025.10		52.13	55.73
VCD 2025.10		50.57	50.66
VisionR1-7B 2026.07		49.5	-
Vanilla 2025.10		49.41	49.45
InternVL2.5-8B 2026.07		49	-
CGD 2025.10		48.89	48.82
AGLA 2025.10		45.2	45.07
CGD 2025.10		33.72	33.34