List your company
ResearchFormerly WizWand

Vision Understanding on MMMU

72.9AccuracyGPT-4o
25.5837.86550.1562.435Jun 3, 2025Aug 2, 2025Oct 2, 2025Dec 2, 2025Feb 1, 2026Apr 3, 2026Jun 3, 2026
Updated 3mo ago

Evaluation results

MethodLinks
GPT-4o2025.0672.9——
LLaVA-1.52025.0667.8——
MetaQuery2025.0658.6——
UniWorld-V12025.0658.6——
BAGEL2025.0655.3——
BLIP3-o2025.0650.6——
DenseModel=Qwen2.5-Omni-7B, Bits=FP162026.0350——
MBQModel=Qwen2.5-Omni-7B, Bits=W8A82026.0350——
MASQuantModel=Qwen2.5-Omni-7B, Bits=W8A82026.0349.9——
RTNModel=Qwen2.5-Omni-7B, Bits=W8A82026.0348.9——
MASQuantModel=Qwen2.5-Omni-7B, Bits=W4A162026.0348.8——
SQModel=Qwen2.5-Omni-7B, Bits=W8A82026.0348.8——
MASQuantModel=Qwen2.5-Omni-7B, Bits=W4A82026.0348.8——
AWQModel=Qwen2.5-Omni-7B, Bits=W4A162026.0347.8——
MBQModel=Qwen2.5-Omni-7B, Bits=W4A162026.0347.8——
MBQModel=Qwen2.5-Omni-7B, Bits=W4A82026.0347.8——
DAVELLM Backbone=Qwen-2.5-7B-Instruct2025.1245.8——
Qwen2.5-Omni (Ours)W/A Bits=4/42026.0645.11——
LaViDa-O2025.1245.1——
FP16 (Full Precision)W/A Bits=16/162026.0644.78——
RTNModel=Qwen2.5-Omni-7B, Bits=W4A162026.0344.4——
SigLIP 2LLM Backbone=Qwen-2.5-7B-Instruct2025.1243.9——
Sparse-LaViDa2025.1243.6——
QLoRAW/A Bits=4/162026.0643.56——
DenseModel=Qwen2.5-Omni-3B, Bits=FP162026.0343.3——
MASQuantModel=Qwen2.5-Omni-3B, Bits=W8A82026.0342.3——
RTNModel=Qwen2.5-Omni-7B, Bits=W4A82026.0342.2——
SQModel=Qwen2.5-Omni-7B, Bits=W4A82026.0342.2——
PoMQ-ViTW/A Bits=4/42026.0642.11——
MBQModel=Qwen2.5-Omni-3B, Bits=W8A82026.0341.8——
SQModel=Qwen2.5-Omni-3B, Bits=W8A82026.0341.7——
AIMv2LLM Backbone=Qwen-2.5-7B-Instruct2025.1241.3——
RTNModel=Qwen2.5-Omni-3B, Bits=W8A82026.0341.3——
QVLMW/A Bits=4/42026.0640.33——
DophineLLM Backbone=Qwen-2.5-7B-Instruct2025.1240.2——
Naive W4A4W/A Bits=4/42026.0640——
MASQuantModel=Qwen2.5-Omni-3B, Bits=W4A162026.0339.6——
MBQModel=Qwen2.5-Omni-3B, Bits=W4A162026.0338.9——
MAE-WebLLM Backbone=Qwen-2.5-7B-Instruct2025.1238.4——
Pix2structLLM Backbone=Qwen-2.5-7B-Instruct2025.1237.9——
SQModel=Qwen2.5-Omni-7B, Bits=W4A62026.0337.8——
SigLIP 2LLM Backbone=Llama-3.2-3B-Instruct2025.1236.9——
MASQuantModel=Qwen2.5-Omni-7B, Bits=W4A62026.0336.8——
AWQModel=Qwen2.5-Omni-3B, Bits=W4A162026.0336.7——
MASQuantModel=Qwen2.5-Omni-3B, Bits=W4A82026.0336.7——
DAVELLM Backbone=Llama-3.2-3B-Instruct2025.1236.6——
Janus-Pro2025.0636.3——
AIMv2LLM Backbone=Llama-3.2-3B-Instruct2025.1236.3——
SeGroSFine-Tuning=Ours2026.0336——
SFTFine-Tuning=SFT2026.0335.3——
DinoV2LLM Backbone=Llama-3.2-3B-Instruct2025.1235——
RecaFine-Tuning=Reca2026.0334.9——
Base ModelFine-Tuning=w/o SFT2026.0334.7——
Pix2StructLLM Backbone=Llama-3.2-3B-Instruct2025.1233.3——
DophineLLM Backbone=Llama-3.2-3B-Instruct2025.1233.3——
MBQModel=Qwen2.5-Omni-7B, Bits=W4A62026.0333.3——
Video-LLaVA2025.0632.8——
RTNModel=Qwen2.5-Omni-3B, Bits=W4A162026.0332.2——
MASQuantModel=Qwen2.5-Omni-3B, Bits=W4A62026.0332.2——
Emu32025.0631.6——
MBQModel=Qwen2.5-Omni-3B, Bits=W4A62026.0331.4——
Janus2025.0630.5——
SQModel=Qwen2.5-Omni-3B, Bits=W4A82026.0330——
MAE-WebLLM Backbone=Llama-3.2-3B-Instruct2025.1229.9——
RTNModel=Qwen2.5-Omni-3B, Bits=W4A82026.0328.9——
RTNModel=Qwen2.5-Omni-3B, Bits=W4A62026.0328.9——
SQModel=Qwen2.5-Omni-3B, Bits=W4A62026.0328.9——
RTNModel=Qwen2.5-Omni-7B, Bits=W4A62026.0328.9——
DiTLLM Backbone=Llama-3.2-3B-Instruct2025.1228.8——
MBQModel=Qwen2.5-Omni-3B, Bits=W4A82026.0327.8——
Show-o2025.0627.4——
BagelType=Und. and Gen., Params=7B2026.01—55.3—
BAGELType=Unified, # LLM Params=1.5B2026.03—43.2—
BAGELType=Unified (>1.5B), # LLM Params=7B2026.03—55.3—
BAGELModel Category=Unified Understanding & Generation Model2025.10——55.3
BAGEL-1.5BStage=Uni. 1 Stage2025.12——43.2
BAGEL-7BStage=Uni. 1 Stage2025.12——55.3
BAGEL-7BLLM Params=3B, Model Category=Und. and Gen.2026.06——43.2
BLIP-3Type=Und. Only, # LLM Params=4B2026.03—41.1—
BLIP3-o-4BLLM Params=4B, Model Category=Und. and Gen.2026.06——46.6
ChameleonType=Unified (>1.5B), # LLM Params=7B2026.03—28.4—
Chameleon-7bType=Unified (Understanding & Generation), LLM Params=7B2025.03—22.4—
Chameleon-7BLLM Params=7B, Model Category=Und. and Gen.2026.06——28.4
DeepSeek-VL2Type=Und. Only, # LLM Params=4.1B/28B2026.03—51.1—
Emu2-ChatType=Understanding-only, LLM Params=33B2025.03—40.7—
Emu3Model Category=Unified Understanding & Generation Model2025.10——31.6
Emu3-ChatSize=8.0B, Capability=Und. only2025.05—31.6—
Emu3-ChatType=Und. Only, # LLM Params=8B2026.03—31.6—
Emu3-Chat-8BLLM Params=8B, Model Category=Und. Only2026.06——31.6
Gemini-Nano-1Type=Unified (Understanding & Generation), LLM Params=1.8B2025.03—26.3—
Harmon-1.5BLLM Params=1.5B, Model Category=Und. and Gen.2026.06——38.9
IdeficsType=Understanding-only, LLM Params=9B2025.03—18.4—
ILLUMEType=Unified (>1.5B), # LLM Params=7B2026.03—38.2—
InternVL2Type=Und. Only, # LLM Params=1.8B2026.03—34.3—
InternVL2Type=Und. Only, # LLM Params=7B2026.03—49.3—
InternVL2.5Type=Und. Only, # LLM Params=1.8B2026.03—43.6—
InternVL2.5Type=Und. Only, # LLM Params=7B2026.03—56—
InternVL3-1BLLM Params=1B, Model Category=Und. Only2026.06——43.4
InternVL3-2BLLM Params=2B, Model Category=Und. Only2026.06——48.2
InternVL3.5Type=Und. only, Params=4B2026.01—66.6—