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

Massive Multi-discipline Multimodal Understanding on MMMU

65.5Accuracy

DART

Updated 5d ago

Evaluation Results

Method
DART 2025.12	65.5	-	-	-	-	-	-
Debate with Consensus 2025.12	63.1	-	-	-	-	-	-
Debate with Judge 2025.12	63.1	-	-	-	-	-	-
Qwen3-VL-8B-Instruct 2026.02	62.4	-	-	-	-	-	-
Qwen3-VL-8B-Thinking 2026.02	62.3	-	-	-	-	-	-
SAP 2026.02	62.3	-	-	-	-	-	-
DART 2025.12	61.6	-	-	-	-	-	-
Self-Consistency (5-way) 2025.12	60.8	-	-	-	-	-	-
Debate with Consensus 2025.12	60.5	-	-	-	-	-	-
Debate with Judge 2025.12	60.3	-	-	-	-	-	-
Claude-3-Opus 2026.02	59.4	-	-	-	-	-	-
CoT 2025.12	58.8	-	-	-	-	-	-
Gemini-1.5 2026.02	58.5	-	-	-	-	-	-
Debate with Judge 2025.12	58.3	-	-	-	-	-	-
Debate with Consensus 2025.12	57	-	-	-	-	-	-
GPT-4V 2026.02	56.8	-	-	-	-	-	-
Self-Refinement 2025.12	56.5	-	-	-	-	-	-
Debate with Judge 2025.12	54.8	-	-	-	-	-	-
Qwen2-VL 2024.11	54.1	-	-	-	-	-	-
ViperGPT 2025.12	54	-	-	-	-	-	-
Grok-1.5V 2026.02	53.6	-	-	-	-	-	-
Debate with Consensus 2025.12	53.2	-	-	-	-	-	-
Full Cache 2025.11	53.18	-	-	-	-	-	-
H2O 2025.11	53.18	-	-	-	-	-	-
SnapKV 2025.11	53.18	-	-	-	-	-	-
LOOK-M 2025.11	53.18	-	-	-	-	-	-
MEDA 2025.11	53.18	-	-	-	-	-	-
FlashCache 2025.11	53.18	-	-	-	-	-	-
Claude-3-Sonnet 2026.02	53.1	-	-	-	-	-	-
Self-Consistency (5-way) 2025.12	52.8	-	-	-	-	-	-
Self-Consistency (5-way) 2025.12	52.5	-	-	-	-	-	-
FlashCache 2025.11	52.27	-	-	-	-	-	-
SnapKV 2025.11	52.18	-	-	-	-	-	-
LOOK-M 2025.11	52.18	-	-	-	-	-	-
Chameleon 2025.12	51.6	-	-	-	-	-	-
MEDA 2025.11	51.36	-	-	-	-	-	-
H2O 2025.11	51.18	-	-	-	-	-	-
Qwen2.5-VL 7B, Dense 2026.03	50.78	-	-	-	-	-	-
CoT 2025.12	50.7	-	-	-	-	-	-
CoT 2025.12	50.4	-	-	-	-	-	-
StreamingLLM 2025.11	50	-	-	-	-	-	-
StreamingLLM 2025.11	50	-	-	-	-	-	-
Gemma3-27B 2025.12	49.7	-	-	-	-	-	-
LLaVA-OneVision 7B, Dense 2026.03	49.22	-	-	-	-	-	-
Self-Refinement 2025.12	47.2	-	-	-	-	-	-
Self-Refinement 2025.12	46.7	-	-	-	-	-	-
Gemma3-4B + AuditDM 2025.12	45.2	-	-	-	-	-	-
LWM 2024.11	44.8	-	-	-	-	-	-
Gemma3-12B 2025.12	44.8	-	-	-	-	-	-
PSFT 2025.08	43.33	-	-	-	-	-	-
SmolVLM2-2.2B 2024.12	42	-	-	-	-	-	-
Base 2025.08	41.89	-	-	-	-	-	-
FlorenceVL-3B 2024.12	41.8	-	-	-	-	-	-
Self-Consistency (5-way) 2025.12	41.3	-	-	-	-	-	-
SFT 2025.08	40.78	-	-	-	-	-	-
PSFT 2025.08	40.78	-	-	-	-	-	-
ATV-Pruning 2026.03	40.11	-	-	-	-	-	-
Base 2025.08	40	-	-	-	-	-	-
ATV-Pruning 2026.03	39.89	-	-	-	-	-	-
Self-Refinement 2025.12	39.7	-	-	-	-	-	-
TAMP 2026.03	39.67	-	-	-	-	-	-
Gemma3-4B 2025.12	39.6	-	-	-	-	-	-
CoT 2025.12	39.4	-	-	-	-	-	-
TAMP 2026.03	39.22	-	-	-	-	-	-
SFT 2025.08	38.78	-	-	-	-	-	-
VisionZip ‡ 2024.12	38.1	-	105.2	-	-	-	-
VisionZip ✨ 2025.08	38.1	-	-	-	-	-	-
FastVLM-1.7B 2024.12	38	-	-	-	-	-	-
VisionZip 2024.12	38	-	-	108.3	-	-	-
VisionZip 2025.08	37.9	-	104.4	-	-	-	-
Wanda 2026.03	37.89	-	-	-	-	-	-
DivPrune 2025.08	37.78	-	-	-	-	-	-
VisionZip 2024.12	37.7	-	-	107.4	-	-	-
SparseGPT 2026.03	37.33	-	-	-	-	-	-
VisionZip 🔥 2025.08	37.3	-	102.8	-	-	-	-
VisionZip 2024.12	37.2	-	-	106	-	-	-
PRUNESID-Dyn 2026.03	37.2	-	-	-	-	-	-
MMTok 2025.08	37.11	-	-	-	-	-	-
MMTok 2025.08	37.11	-	-	-	-	-	-
VisionZip 2024.12	37	-	102.2	-	-	-	-
VisionZip 2025.08	37	-	-	-	-	-	-
VisionZip ‡ 2024.12	36.9	-	101.9	-	-	-	-
VisionZip ✨ 2025.08	36.9	-	-	-	-	-	-
PRUNESID-Dyn 2026.03	36.8	-	-	-	-	-	-
DreamLLM+ 2024.11	36.6	-	-	-	-	-	-
VisionZip 2025.08	36.6	-	100.8	-	-	-	-
VisionZip 2024.12	36.4	-	100.5	-	-	-	-
VisionZip 2025.08	36.4	-	-	-	-	-	-
MMTok 2025.08	36.33	-	100.1	-	-	-	-
MMTok 2025.08	36.33	-	-	-	-	-	-
LLaVA-1.5 Vanilla 7B 2025.08	36.3	-	100	-	-	-	-
Vanilla 2026.03	36.3	-	-	-	-	-	-
Vanilla 13B 2024.12	36.2	-	100	-	-	-	-
VisionZip 2024.12	36.2	-	-	-	100.3	-	-
VisionZip 🔥 2025.08	36.2	-	99.7	-	-	-	-
VisionZip 2025.08	36.2	-	99.7	-	-	-	-
LLaVA-NeXT Vanilla 13B 2025.08	36.2	-	-	-	-	-	-
MMTok 2025.08	36.11	-	99.5	-	-	-	-
VisionZip 2024.12	36.1	-	-	102.8	-	-	-
Vanilla 7B 2024.12	36.1	-	-	-	100	-	-

Showing 100 of 214 rows