Visual Question Answering on Super-CLEVR (test) benchmark leaderboard · SOTA2 Research

Visual Question Answering on Super-CLEVR (test)

87.4AccuracyAtt-CoT

Evaluation Results

Method	Links
Att-CoT2026.06		87.4	63.2	0.231	0.52	-1.22
Att-CoT2026.06		86.5	57.1	0.643	0.43	-0.58
Att-CoT2026.06		86.3	62.9	0.347	0.32	-1.64
SFT2026.06		85.5	60.4	0.478	0.35	0.75
SFT2026.06		84.2	54.8	0.897	0.41	1.45
LLaVA-R2026.06		83.9	61.1	0.36	0.35	—
Att-CoT2026.06		83.7	56.9	0.32	0.18	-1.24
LLaVA-CoT2026.06		83.2	58.9	0.412	0.31	—
Att-CoT2026.06		83.1	56.5	0.806	0.55	1.82
LLaVA-R2026.06		82.8	54.5	0.812	0.4	—
SFT2026.06		82.1	54.2	0.515	-0.03	0.33
LLaVA-R2026.06		82	54.6	0.464	0.08	—
SFT2026.06		81.8	60.6	0.502	0.11	3.58
LLaVA-R2026.06		81.4	61.2	0.438	0.22	—
LLaVA-CoT2026.06		80.6	53.1	0.527	-0.01	—
LLaVA-CoT2026.06		80.6	59.3	0.516	0.09	—
LLaVA-CoT2026.06		80.2	52.9	0.876	0.38	—
SFT2026.06		79.8	55.1	0.882	0.45	5.59
LLaVA-R2026.06		78.7	55.2	0.842	0.4	—
LLaVA-CoT2026.06		76.4	54.1	0.819	0.36	—
Att-CoT2026.06		74.5	54.7	0.754	0.71	-0.2
SFT2026.06		65.5	52.6	0.925	0.65	8.86
LLaVA-R2026.06		64.9	52.8	0.796	0.62	—
LLaVA-CoT2026.06		62.8	49.4	0.855	0.6	—