Qwen2.5 VL 72B Instruct
Alibaba Cloud / Qwen TeamQwen开源权重tongyi-qianwen
描述
Qwen2.5-VL is the new flagship vision-language model of Qwen, significantly improved from Qwen2-VL. It excels at recognizing objects, analyzing text/charts/layouts in images, acting as a visual agent, understanding long videos (over 1 hour) with event pinpointing, performing visual localization (bounding boxes/points), and generating structured outputs from documents.
发布日期
2025-01-26
参数规模
72.0B
上下文长度
131K
支持模态
image, text
能力雷达图
70
general
0
coding
60
reasoning
60
science估算
40
agents
80
multimodal
缺少专门科学评测时，Science 由 LLM Stats 科学得分或推理能力估算。
排行榜排名
| 领域 | #排名 | 分数 | 来源 |
|---|---|---|---|
| 多模态榜 | 137 | 29.0 | LS |
基准测试分数 (LLM Stats)
(LLM Stats (zeroeval))General
MLVU-M
74.6%自报
AndroidWorld_SR
35.0%自报
Math
MathVista-Mini
74.8%自报
MathVision
38.1%自报
Multimodal
Android Control Low_EM
93.7%自报
AITZ_EM
83.2%自报
VideoMME w/o sub.
73.3%自报
MMMU
70.2%自报
MobileMiniWob++_SR
68.0%自报
Android Control High_EM
67.4%自报
OSWorld
8.8%自报
Reasoning
ChartQAMasry et al. (2022)
89.5%自报
Video
PerceptionTest
73.2%自报
MMBench-Video
2.0%自报
Vision
DocVQADocVQA (2020)
96.4%自报
OCRBench
88.5%自报
AI2D
88.4%自报
MMBench
88.0%自报
ScreenSpot
87.1%自报
CC-OCR
79.8%自报
EgoSchema
76.2%自报
MMVet
76.2%自报
TempCompass
74.8%自报
MMStar
70.8%自报
MVBench
70.4%自报
OCRBench-V2 (en)
61.5%自报
Hallusion Bench
55.2%自报
MMMU-Pro
51.1%自报
LVBench
47.3%自报
ScreenSpot Pro
43.6%自报
AA 评测指数
(Artificial Analysis)暂无 AA 评测数据
LLM Stats 分类评分
(LLM Stats (zeroeval))Image To Text80
Structured Output80
Text-to-image80
Multimodal70
Reasoning70
Spatial Reasoning70
General70
Grounding70
Healthcare70
Vision70
Long Context60
Math60
Video50
Agents40
定价
输入价格$2.8 / 1M tokens
输出价格$8.4 / 1M tokens
混合价格(3:1)$4.2 / 1M tokens
速度
暂无速度数据
供应商价格排行
供应商价格排行
6 个供应商
最便宜: OpenRouter最贵: Alibaba
供应商输入输出
1OpenRouter最便宜
$0.8
$1
2NovitaAI
$0.8
$0.8
3Kilo Gateway
$0.8
$1
4Alibaba (China)
$2.294
$6.881
5Alibaba Cloud / Qwen Team主要
$2.8
$8.4
6Alibaba
$2.8
$8.4
比较该模型在不同 API 供应商之间的定价。