Qwen2.5 VL 72B Instruct
Alibaba Cloud / Qwen TeamQwen開源權重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.
發布日期
2025-01-26
參數規模
72.0B
上下文長度
131K
支援模態
image, text
能力雷達圖
70
general
0
coding
60
reasoning
60
science估算
40
agents
80
multimodal
缺少專門科學評測時，Science 由 LLM Stats 科學得分或推理能力估算。
排行榜排名
基準測試分數 (LLM Stats)
(LLM Stats (zeroeval))Agents
AITZ_EM
83.2%自報
MobileMiniWob++_SR
68.0%自報
AndroidWorld_SR
35.0%自報
OSWorld
8.8%自報
General
MMVet
76.2%自報
MLVU-M
74.6%自報
MMStar
70.8%自報
MMMU
70.2%自報
MMMU-Pro
51.1%自報
Grounding
ScreenSpot
87.1%自報
ScreenSpot Pro
43.6%自報
Image To Text
DocVQADocVQA (2020)
96.4%自報
OCRBench
88.5%自報
OCRBench-V2 (en)
61.5%自報
Long Context
EgoSchema
76.2%自報
LVBench
47.3%自報
Math
MathVista-Mini
74.8%自報
MathVision
38.1%自報
Multimodal
Android Control Low_EM
93.7%自報
ChartQAMasry et al. (2022)
89.5%自報
AI2D
88.4%自報
MMBench
88.0%自報
CC-OCR
79.8%自報
TempCompass
74.8%自報
VideoMME w/o sub.
73.3%自報
PerceptionTest
73.2%自報
MVBench
70.4%自報
Android Control High_EM
67.4%自報
MMBench-Video
2.0%自報
Reasoning
Hallusion Bench
55.2%自報
AA 評測指數
(Artificial Analysis)暫無 AA 評測資料
LLM Stats 分類評分
(LLM Stats (zeroeval))Structured Output80
Image To Text80
Text-to-image80
Reasoning70
Spatial Reasoning70
General70
Grounding70
Healthcare70
Long Context60
Math60
Multimodal60
Vision60
Video50
Agents40
定價
輸入價格$2.8 / 1M tokens
輸出價格$8.4 / 1M tokens
混合價格(3:1)$4.2 / 1M tokens
速度
暫無速度資料
供應商價格排行
供應商價格排行
7 個供應商
最便宜: OpenRouter最貴: Alibaba
供應商輸入輸出
1OpenRouter最便宜
$0.25
$0.75
2Kilo Gateway
$0.25
$0.75
3LLM Gateway
$0.25
$0.75
4NovitaAI
$0.8
$0.8
5Alibaba (China)
$2.294
$6.881
6Alibaba Cloud / Qwen Team主要
$2.8
$8.4
7Alibaba
$2.8
$8.4
比較該模型在不同 API 供應商之間的定價。