List your company
ResearchFormerly WizWand

Multi-modal Vision-Language Understanding on GQA

64.2AccuracyLLaVA-NeXT
37.36844.33451.358.266May 23, 2025Jul 20, 2025Sep 16, 2025Nov 14, 2025Jan 11, 2026Mar 10, 2026May 8, 2026
Updated 5mo ago

Evaluation results

MethodLinks
LLaVA-NeXTLLM=Vicuna-7B2025.0564.2——
TUNAType=Native Unified, # Params.=7B2026.0563.9——
ASPOBase Model=LLaVA-v1.5, Model Scale=13B, Optimization Protocol=ASPO2025.0563.468.8—
LLaVA-1.5-13BBase Model=LLaVA-v1.5, Model Scale=13B, Optimization Protocol=Base2025.0563.365.93—
Show-o2Params=7B, Architecture category=Integrating Autoregressive and Diffusion within Transformers2026.0363.1——
Show-o2Type=Native Unified, # Params.=7B2026.0563.1——
LLaVA-NeXT 7BToken Budget=2880, Base Model=LLaVA-NeXT 7B2026.0362.8—100
TokenFlowParams=13B, Architecture category=Unifying Understanding and Generation via Next-token Prediction2026.0362.7——
DPOBase Model=LLaVA-v1.5, Model Scale=13B, Optimization Protocol=DPO2025.0562.367.03—
LLaVA-1.5-7BBase Model=LLaVA-v1.5, Model Scale=7B, Optimization Protocol=Base2025.056262.59—
ASPOBase Model=LLaVA-v1.5, Model Scale=7B, Optimization Protocol=ASPO2025.056265.16—
Janus-ProParams=7B, Architecture category=Unifying Understanding and Generation via Next-token Prediction2026.0362——
LLaVA-v1.5Type=Und. Only, # Params.=7B2026.0562——
Janus-ProType=Native Unified, # Params.=7B2026.0562——
PromPruneToken Budget=640, Base Model=LLaVA-NeXT 7B2026.0361.8—99.6
DivPruneToken Budget=640, Base Model=LLaVA-NeXT 7B2026.0361.7—99.6
TUNAType=Native Unified, # Params.=1.5B2026.0561.4——
VisPrunerToken Budget=640, Base Model=LLaVA-NeXT 7B2026.0361.3—99.2
DPOBase Model=LLaVA-v1.5, Model Scale=7B, Optimization Protocol=DPO2025.056163.12—
DivPruneToken Budget=320, Base Model=LLaVA-NeXT 7B2026.0361—97.7
MogaoParams=7B, Architecture category=Integrating Autoregressive and Diffusion within Transformers2026.0360.9——
MogaoType=Native Unified, # Params.=7B2026.0560.9——
VILA-UParams=7B, Architecture category=Unifying Understanding and Generation via Next-token Prediction2026.0360.8——
PromPruneToken Budget=320, Base Model=LLaVA-NeXT 7B2026.0360.7—97.9
Qwen-2.5-VL-InstructType=Und. Only, # Params.=7B2026.0560.7——
JanusFlowParams=1.3B, Architecture category=Integrating Autoregressive and Diffusion within Transformers2026.0360.3——
Emu3Params=8B, Architecture category=Unifying Understanding and Generation via Next-token Prediction2026.0360.3——
Qwen2.5 VLParams=7B, Architecture category=Unifying Understanding and Generation via Next-token Prediction, VLMEvalKit=true2026.0360.3——
Emu3Type=Native Unified, # Params.=8B2026.0560.3——
WallarooParams=7B, Architecture category=Unifying Understanding and Generation via Next-token Prediction, VLMEvalKit=true2026.0360.1——
DivPruneToken Budget=160, Base Model=LLaVA-NeXT 7B2026.0359.5—95
Qwen-VL-7BBase Model=Qwen-VL, Model Scale=7B, Optimization Protocol=Base2025.0559.3——
JanusParams=1.5B, Architecture category=Unifying Understanding and Generation via Next-token Prediction2026.0359.1——
PromPruneToken Budget=160, Base Model=LLaVA-NeXT 7B2026.0359.1—95.2
Slot-MLLM (14B)LLM=Qwen2.5-14B-Instruct2025.0558.8——
VisPrunerToken Budget=320, Base Model=LLaVA-NeXT 7B2026.0358.7—95.9
Show-oParams=1.3B, Architecture category=Integrating Autoregressive and Diffusion within Transformers2026.0358——
Qwen-VL-chat-7BBase Model=Qwen-VL-Chat, Model Scale=7B, Optimization Protocol=Base2025.0557.5——
Qwen-VL-ChatType=Und. Only, # Params.=7B2026.0557.5——
Slot-MLLM (7B)LLM=Vicuna-7B2025.0557.3——
mPLUG-Owl2-7BBase Model=mPLUG-Owl2, Model Scale=7B, Optimization Protocol=Base2025.0556.1——
STARFlow2Type=Native Unified, # Params.=10.6B2026.0555.8——
VisPrunerToken Budget=160, Base Model=LLaVA-NeXT 7B2026.0355.6—90.6
InstructBLIPLLM=Vicuna-7B2025.0549.2——
SEED-XType=Composite Unified, # Params.=17B2026.0549.1——
ASPOBase Model=InstructBLIP, Model Scale=13B, Optimization Protocol=ASPO2025.0548.245.84—
InstructBLIP-13BBase Model=InstructBLIP, Model Scale=13B, Optimization Protocol=Base2025.0548.143.86—
DPOBase Model=InstructBLIP, Model Scale=13B, Optimization Protocol=DPO2025.0547.544.5—
IDEFICS-65BBase Model=IDEFICS, Model Scale=65B, Optimization Protocol=Base2025.0545.2——
BLIP-2-7BBase Model=BLIP-2, Model Scale=7B, Optimization Protocol=Base2025.0541——
IDEFICS-7BBase Model=IDEFICS, Model Scale=7B, Optimization Protocol=Base2025.0538.4——