DeepSeek VL2
Multimodal
Zero-eval
#1MMT-Bench
#1MME
#3MMBench-V1.1
+1 more
by DeepSeek
+
+
+
+
About
DeepSeek-VL2 was developed as a vision-language model, designed to handle both visual and textual inputs for multimodal understanding tasks. Built to extend DeepSeek's capabilities beyond text-only processing, it enables applications requiring integrated analysis of images and language.
+
+
+
+
Pricing Range
Input (per 1M)$9.50 -$9.50
Output (per 1M)$4800.00 -$4800.00
Providers1
+
+
+
+
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
70.9%
Best Score
93.3%
High Performers (80%+)
5Performance Metrics
Max Context Window
258.6KAvg Throughput
22.0 tok/sAvg Latency
1ms+
+
+
+
All Benchmark Results for DeepSeek VL2
Complete list of benchmark scores with detailed information
| DocVQA | multimodal | 0.93 | 93.3% | Self-reported | |
| ChartQA | multimodal | 0.86 | 86.0% | Self-reported | |
| TextVQA | multimodal | 0.84 | 84.2% | Self-reported | |
| AI2D | multimodal | 0.81 | 81.4% | Self-reported | |
| OCRBench | multimodal | 0.81 | 81.1% | Self-reported | |
| MMBench | multimodal | 0.80 | 79.6% | Self-reported | |
| MMBench-V1.1 | multimodal | 0.79 | 79.2% | Self-reported | |
| InfoVQA | multimodal | 0.78 | 78.1% | Self-reported | |
| RealWorldQA | multimodal | 0.68 | 68.4% | Self-reported | |
| MMT-Bench | multimodal | 0.64 | 63.6% | Self-reported |
Showing 1 to 10 of 14 benchmarks
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+