List your company
ResearchFormerly WizWand

MM 2D Grounding on CountBenchQA

93.8ScoreQwen3-VL-32B
35.643250.741665.8480.9384Jun 23, 2026Jun 25, 2026Jun 27, 2026Jun 29, 2026Jul 1, 2026Jul 3, 2026Jul 5, 2026
Updated 3mo ago

Evaluation results

MethodLinks
Qwen3-VL-32BDecoding Mode=Instruct, Model Scale=32B2026.0693.8
K2.5Parameter Count=≈1T2026.0693.8
Yuvion VL-32BDecoding Mode=Instruct, Model Scale=32B2026.0693
Qwen3.5-PlusParameter Count=397B2026.0692.8
Qwen3-VL-32BDecoding Mode=Thinking, Model Scale=32B2026.0692.2
Qwen3-VL-8BDecoding Mode=Instruct, Model Scale=8B2026.0690.8
Yuvion VL-8BDecoding Mode=Instruct, Model Scale=8B2026.0689.9
Yuvion VL-32BDecoding Mode=Reasoning, Model Scale=32B2026.0689.5
Opus-4.6Parameter Count=≈800B2026.0688.7
Yuvion VL-8BDecoding Mode=Reasoning, Model Scale=8B2026.0688.3
GPT-5.4Parameter Count=≈1.5T2026.0688.3
Qwen3-VL-8BFine-tuning setting=Base2026.0786.97
Qwen3-VL-8BDecoding Mode=Thinking, Model Scale=8B2026.0686.9
Qwen3-VL-8BFine-tuning setting=CPO2026.0785.74
BaseFine-tuning setting=Base, Backbone=Qwen3-VL-2B2026.0783.3
BaseModel=Qwen3-VL-4B2026.0782.89
LoRAModel=Qwen3-VL-4B2026.0780.24
Qwen3-VL-8BFine-tuning setting=GSPO2026.0779.63
GSPOModel=Qwen3-VL-4B2026.0776.17
LoRAFine-tuning setting=LoRA, Backbone=Qwen3-VL-2B2026.0771.28
Qwen3-VL-8BFine-tuning setting=LoRA2026.0770.67
GRPOModel=Qwen3-VL-4B2026.0770.26
CPOModel=Qwen3-VL-4B2026.0768.64
CPOFine-tuning setting=CPO, Backbone=Qwen3-VL-2B2026.0765.99
Qwen3-VL-8BFine-tuning setting=GRPO2026.0764.15
Qwen3-VL-8BFine-tuning setting=FFT2026.0763.14
FFTFine-tuning setting=FFT, Backbone=Qwen3-VL-2B2026.0762.32
FFTModel=Qwen3-VL-4B2026.0761.3
GSPOFine-tuning setting=GSPO, Backbone=Qwen3-VL-2B2026.0760.08
GRPOFine-tuning setting=GRPO, Backbone=Qwen3-VL-2B2026.0737.88