An advanced series of large Mixture-of-Experts (MoE) Vision-Language Models that significantly improves upon its predecessor, DeepSeek-VL. DeepSeek-VL2 demonstrates superior capabilities across various tasks, including but not limited to visual question answering, optical character recognition, document/table/chart understanding, and visual grounding.
发布日期2024年12月13日
参数规模27B
上下文长度—
许可证deepseek
知识截止—
Benchmarks
评测成绩
| 评测基准 | 类别 | 分数 | 来源 |
|---|---|---|---|
| DocVQA | vision multimodal | 93.3 | 来源 |
| ChartQA | reasoning vision multimodal | 86.0 | 来源 |
| TextVQA | vision multimodal image-to-text | 84.2 | 来源 |
| 多模态开源模型榜·AI2D | vision | 83.8 | 来源 |
| AI2D | vision reasoning multimodal | 81.4 | 来源 |
| 多模态开源模型榜·MMBench TEST | vision | 81.2 | 来源 |
| OCRBench | vision image-to-text | 81.1 | 来源 |
| 多模态开源模型榜·OCRBench | vision | 80.9 | 来源 |
| MMBench | vision multimodal reasoning | 79.6 | 来源 |
| MMBench-V1.1 | vision multimodal reasoning | 79.2 | 来源 |
| InfoVQA | vision multimodal | 78.1 | 来源 |
| RealWorldQA | vision spatial_reasoning | 68.4 | 来源 |
| 多模态开源模型榜 | vision | 66.4 | 来源 |
| 多模态开源模型榜·MathVista | vision | 63.9 | 来源 |
| MMT-Bench | vision multimodal reasoning general | 63.6 | 来源 |
| MathVista | math vision multimodal | 62.8 | 来源 |
| 多模态开源模型榜·MMStar | vision | 61.9 | 来源 |
| MMStar | vision multimodal reasoning general | 61.3 | 来源 |
| 多模态开源模型榜·MMVet | vision | 60.0 | 来源 |
| 多模态安全榜·Toxicity | security vision | 55.8 | — |
| 多模态开源模型榜·MMMU VAL | vision | 54.0 | 来源 |
| 多模态安全榜·Privacy | security vision | 51.6 | — |
| MMMU | multimodal reasoning general | 51.1 | 来源 |
| 多模态安全榜·Health Risk | security vision | 49.3 | — |
| 多模态安全榜·Legality | security vision | 48.8 | — |
| 多模态开源模型榜·HallusionBench | vision | 45.3 | 来源 |
| 多模态安全榜 | security vision | 42.7 | — |
| 多模态安全榜·Misinformation | security vision | 30.1 | — |
| MME | vision multimodal reasoning | 22.5 | 来源 |
| 多模态安全榜·Bias | security vision | 20.6 | — |
Pricing
API 价格对比
暂无 API 价格。