List your company
ResearchFormerly WizWand

General Visual Question Answering on RealWorldQA

73.1ScoreBee-8B
42.21250.23158.2566.269Oct 15, 2025Nov 27, 2025Jan 10, 2026Feb 23, 2026Apr 8, 2026May 22, 2026Jul 5, 2026
Updated 3mo ago

Evaluation results

MethodLinks
Bee-8BOpenness=fully open, Model Size=8B, Training Strategy=RL2025.1073.1
MolmoOpenness=fully open, Model Size=7B, Training Strategy=Distilled (-D)2025.1070.7
Bee-8BOpenness=fully open, Model Size=8B, Training Strategy=SFT2025.1070.1
SpatialSSRLBackbone=Qwen2.5-VL-7B2026.0469.9
BaselineBackbone=Qwen2.5-VL-7B2026.0469.5
Qwen2.5-VLOpenness=semi-open, Model Size=7B2025.1068.5
BaseModel=Qwen3-VL-4B2026.0768.24
Keye-VLOpenness=semi-open, Model Size=8B2025.1067.7
InternVL3.5Openness=semi-open, Model Size=8B2025.1067.5
SpatialEvoBackbone=Qwen2.5-VL-7B2026.0466.7
SpatialEvoBackbone=Qwen2.5-VL-3B2026.0466.5
LLaVA OneVisionOpenness=fully open, Model Size=7B2025.1066.3
SpatialSSRLBackbone=Qwen2.5-VL-3B2026.0465.4
SpaceRBackbone=Qwen2.5-VL-7B2026.0464.7
CPOModel=Qwen3-VL-4B2026.0764.58
BaselineBackbone=Qwen2.5-VL-3B2026.0463.4
LoRAModel=Qwen3-VL-4B2026.0761.7
SpaceRBackbone=Qwen2.5-VL-3B2026.0461.4
GSPOModel=Qwen3-VL-4B2026.0758.17
HIVEBackbone=Huginn, Param.=4B, Train Data=6.5M, Recurrence Level (r)=322026.0257.9
ViLaSRBackbone=Qwen2.5-VL-7B2026.0457.9
BaseFine-tuning setting=Base, Backbone=Qwen3-VL-2B2026.0757.78
Emu3Param.=8B2026.0257.4
SpatialLadderBackbone=Qwen2.5-VL-3B2026.0457.1
CPOFine-tuning setting=CPO, Backbone=Qwen3-VL-2B2026.0754.77
GRPOFine-tuning setting=GRPO, Backbone=Qwen3-VL-2B2026.0754.51
LoRAFine-tuning setting=LoRA, Backbone=Qwen3-VL-2B2026.0754.25
FFTModel=Qwen3-VL-4B2026.0751.63
FFTFine-tuning setting=FFT, Backbone=Qwen3-VL-2B2026.0751.11
GSPOFine-tuning setting=GSPO, Backbone=Qwen3-VL-2B2026.0746.01
Gemma3-4B-PTBackbone=Gemma3, Param.=4B, Train Data=4T*2026.0245.5
GRPOModel=Qwen3-VL-4B2026.0743.4