List your company

OCR Benchmarking on OCRBench v2 (en/cn metrics)

62.6AccuracySigLIP2-g
13.30426.10238.951.698Jul 8, 2025Sep 6, 2025Nov 5, 2025Jan 4, 2026Mar 5, 2026May 4, 2026Jul 3, 2026

Evaluation results

MethodLinks
SigLIP2-gTokens per frame/tile=256, TTFT (ms)=517.6, LLM Backbone=9B Nemotron2026.0762.6———
RADIO1DTokens per frame/tile=128, TTFT (ms)=373.2, LLM Backbone=9B Nemotron2026.0761.5———
RADIO1DTokens per frame/tile=256, TTFT (ms)=452.8, LLM Backbone=9B Nemotron2026.0761.4———
SigLIP2-SO400mTokens per frame/tile=256, TTFT (ms)=440.5, LLM Backbone=9B Nemotron2026.0761.1———
RADIO1DTokens per frame/tile=192, TTFT (ms)=380.4, LLM Backbone=9B Nemotron2026.0760.9———
C-RADIOv4-HTokens per frame/tile=256, TTFT (ms)=468.2, LLM Backbone=9B Nemotron2026.0760.4———
RADIO1DTokens per frame/tile=224, TTFT (ms)=410.2, LLM Backbone=9B Nemotron2026.0760.2———
RADIO1DTokens per frame/tile=64, TTFT (ms)=346.1, LLM Backbone=9B Nemotron2026.0759.5———
RADIO1DTokens per frame/tile=32, TTFT (ms)=335.1, LLM Backbone=9B Nemotron2026.0755.4———
RADIO1DTokens per frame/tile=8, TTFT (ms)=329.7, LLM Backbone=9B Nemotron2026.0749.2———
RADIO1DTokens per frame/tile=256, TTFT (ms)=452.8, LLM Backbone=9B Nemotron2026.0742.2———
RADIO1DTokens per frame/tile=224, TTFT (ms)=410.2, LLM Backbone=9B Nemotron2026.0741.8———
C-RADIOv4-HTokens per frame/tile=256, TTFT (ms)=468.2, LLM Backbone=9B Nemotron2026.0741.5———
SigLIP2-gTokens per frame/tile=256, TTFT (ms)=517.6, LLM Backbone=9B Nemotron2026.0741———
SigLIP2-SO400mTokens per frame/tile=256, TTFT (ms)=440.5, LLM Backbone=9B Nemotron2026.0740.7———
RADIO1DTokens per frame/tile=192, TTFT (ms)=380.4, LLM Backbone=9B Nemotron2026.0740.2———
RADIO1DTokens per frame/tile=128, TTFT (ms)=373.2, LLM Backbone=9B Nemotron2026.0738.6———
RADIO1DTokens per frame/tile=64, TTFT (ms)=346.1, LLM Backbone=9B Nemotron2026.0735.2———
RADIO1DTokens per frame/tile=1, TTFT (ms)=327.0, LLM Backbone=9B Nemotron2026.0734.1———
RADIO1DTokens per frame/tile=32, TTFT (ms)=335.1, LLM Backbone=9B Nemotron2026.0734.1———
PAPO_GSize=7B, Base model=Qwen2.5-VL2025.0732.11———
GRPOSize=7B, Base model=Qwen2.5-VL2025.0731.17———
RADIO1DTokens per frame/tile=8, TTFT (ms)=329.7, LLM Backbone=9B Nemotron2026.0727.9———
BaseSize=7B, Base model=Qwen2.5-VL2025.0727.75———
PAPO_GSize=3B, Base model=Qwen2.5-VL2025.0727.22———
GRPOSize=3B, Base model=Qwen2.5-VL2025.0726.32———
BaseSize=3B, Base model=Qwen2.5-VL2025.0720.14———
RADIO1DTokens per frame/tile=1, TTFT (ms)=327.0, LLM Backbone=9B Nemotron2026.0715.2———
AODModel=LLaVA-1.5-7B2026.05———27.2
AODModel=Qwen2.5-VL-7B2026.05———52.4
AODModel=InternVL3-8B2026.05———54.4
ASDModel=LLaVA-1.5-7B2026.05———23.6
ASDModel=Qwen2.5-VL-7B2026.05———49.2
ASDModel=InternVL3-8B2026.05———51.5
BaseModel=LLaVA-1.5-7B2026.05———16.8
BaseModel=Qwen2.5-VL-7B2026.05———46.7
BaseModel=InternVL3-8B2026.05———49
Gemini-2.5-flash2025.09—5657.9—
GPT-4o-2024-05-132025.09—55.857.6—
LaV-CoTModel Scale=3B, Training/Optimization Strategy=SFT2025.09—55.257.6—
LaV-CoTModel Scale=3B, Training/Optimization Strategy=SFT+GRPO2025.09—56.458.2—
PruneHalModel=LLaVA-1.5-7B2026.05———20.2
PruneHalModel=Qwen2.5-VL-7B2026.05———48.9
PruneHalModel=InternVL3-8B2026.05———52.1
Qwen2.5-VLModel Scale=3B2025.09—52.154.3—
Qwen2.5-VLModel Scale=7B2025.09—53.256.1—
TruthPrIntModel=LLaVA-1.5-7B2026.05———25.1
TruthPrIntModel=Qwen2.5-VL-7B2026.05———50.1
TruthPrIntModel=InternVL3-8B2026.05———52.9
VASparseModel=LLaVA-1.5-7B2026.05———17.2
VASparseModel=Qwen2.5-VL-7B2026.05———48.5
VASparseModel=InternVL3-8B2026.05———48.7
VCDModel=LLaVA-1.5-7B2026.05———19.4
VCDModel=Qwen2.5-VL-7B2026.05———47.8
VCDModel=InternVL3-8B2026.05———49.1