DeepSeek

DeepSeek VL2 Tiny

Multimodal
Zero-eval
#3MME

by DeepSeek

+
+
+
+
About

DeepSeek-VL2-Tiny was developed as an ultra-efficient vision-language model, designed for deployment in resource-constrained environments. Built to enable multimodal AI on edge devices and mobile applications, it distills vision-language capabilities into a minimal footprint for widespread accessibility.

+
+
+
+
Timeline
AnnouncedDec 13, 2024
ReleasedDec 13, 2024
+
+
+
+
Specifications
Capabilities
Multimodal
+
+
+
+
License & Family
License
deepseek
Performance Overview
Performance metrics and category breakdown

Overall Performance

14 benchmarks
Average Score
63.1%
Best Score
88.9%
High Performers (80%+)
4
+
+
+
+
All Benchmark Results for DeepSeek VL2 Tiny
Complete list of benchmark scores with detailed information
DocVQA
multimodal
0.89
88.9%
Self-reported
ChartQA
multimodal
0.81
81.0%
Self-reported
OCRBench
multimodal
0.81
80.9%
Self-reported
TextVQA
multimodal
0.81
80.7%
Self-reported
AI2D
multimodal
0.72
71.6%
Self-reported
MMBench
multimodal
0.69
69.2%
Self-reported
MMBench-V1.1
multimodal
0.68
68.3%
Self-reported
InfoVQA
multimodal
0.66
66.1%
Self-reported
RealWorldQA
multimodal
0.64
64.2%
Self-reported
MathVista
multimodal
0.54
53.6%
Self-reported
Showing 1 to 10 of 14 benchmarks
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+