跳轉到主要內容

Qwen3 VL 4B (Reasoning)

AlibabaQwen開源權重Apache 2.0 · 商用許可

描述

Qwen3-VL is a large multimodal model that unifies vision, language, and reasoning to achieve human-level perception and cognition across text, images, and video. Built on a 235B-parameter architecture, it integrates early joint training of visual and textual modalities for strong language grounding. The model supports up to a 1 million-token context window and excels at visual understanding, spatial reasoning, long video comprehension, and tool-based interaction. It can generate code from images, perform precise 2D/3D object grounding, and operate digital interfaces like a visual agent. The “Instruct” version rivals Gemini 2.5 Pro in perception benchmarks, while the “Thinking” version leads in multimodal reasoning and STEM tasks. With multilingual OCR, creative writing, and fine-grained scene interpretation, Qwen3-VL establishes a new open-source frontier for integrated vision-language intelligence.

發布日期
2025-10-14
參數規模
4.0B
上下文長度
—
支援模態
image, text

能力雷達圖

26
general
32
coding
30
reasoning
36
science
70
agents
100
multimodal

排行榜排名

領域#排名分數來源
程式碼能力榜506
20.0
AA
通用能力榜498
28.0
AA
多模態榜140
25.0
LS
科學能力513
26.0
AA

基準測試分數 (LLM Stats)

(LLM Stats (zeroeval))

Chat

MM-MT-Bench7.70 / 100自報
IFEvalGoogle Research (2023)82.6%自報
Multi-IF73.6%自報

Creativity

Creative Writing v376.1%自報

General

MMLU81.5%自報
MLVU-M75.7%自報
BFCL-v367.3%自報
Include64.6%自報
Arena-Hard v236.8%自報

Language

MMLU-Redux86.0%自報
MMLU-Pro73.6%自報
MMLU-ProX65.0%自報

Math

MathVista-Mini79.5%自報
AIME 202574.5%自報
LiveBench 2024112568.4%自報
MathVision60.0%自報
HMMT2553.1%自報
PolyMATH44.6%自報

Multimodal

VideoMMMU69.4%自報
OSWorld31.4%自報

Reasoning

CharXiv-D83.9%自報
GPQANYU + Cohere + Anthropic (2023)64.1%自報
LiveCodeBench v651.3%自報
CharXiv-R50.3%自報
SuperGPQA46.8%自報

Video

CharadesSTA59.0%自報

Vision

DocVQAtest94.2%自報
ScreenSpot92.9%自報
MMBench-V1.186.7%自報
AI2D84.9%自報
InfoVQAtest83.0%自報
OCRBench80.8%自報
MuirBench75.0%自報
CC-OCR73.8%自報
MMStar73.2%自報
RealWorldQA73.2%自報
MMMU (val)70.8%自報
MVBench69.3%自報
Hallusion Bench64.1%自報
BLINK63.4%自報
OCRBench-V2 (en)61.8%自報
MMMU-Pro57.0%自報
OCRBench-V2 (zh)55.8%自報
LVBench53.5%自報
ScreenSpot Pro49.2%自報
ERQA47.3%自報
ODinW39.4%自報

Writing

WritingBench84.0%自報

AA 評測指數

(Artificial Analysis)
Mmlu Pro(TIGER-Lab (Univ. of Waterloo, Toronto, CMU, 2024))
70.0
Gpqa(NYU + Cohere + Anthropic (2023))
49.4
Ifbench(Google Research (2023))
36.6
Livecodebench(UC Berkeley + MIT + Cornell (2024))
32.0
Math Index(Artificial Analysis)
25.7
Aime 25(MAA (Mathematical Association of America))
25.7
Lcr(Artificial Analysis)
21.3
Tau2(Sierra + U Toronto + Vector Institute (2025))
15.5
Intelligence Index(Artificial Analysis)
7.0
Hle(Center for AI Safety + Scale AI (2025))
4.6
Terminalbench Hard(Stanford × Laude Institute (2026))
1.5

LLM Stats 分類評分

(LLM Stats (zeroeval))
Chat
3
Communication
3
Multimodal
100
Instruction Following
80
Structured Output
80
General
80
Image To Text
70
Language
70
Legal
70
Math
70
Reasoning
70
Finance
70
Grounding
70
Healthcare
70
Creativity
70
Text-to-image
70
Tool Calling
70
Vision
70
Writing
70
Physics
60
Spatial Reasoning
60
3d
60
Biology
60
Chemistry
60
Video
60
Long Context
50
Agents
50
Economics
50

定價

輸入價格免費
輸出價格免費
混合價格(3:1)免費

速度

Tokens/秒0.0
首Token延遲0.00s
首回答延遲0.00s

供應商價格排行

暫無提供商資料

外部連結