跳转到主要内容

Qwen3 VL 4B Instruct

AlibabaQwen开源权重Apache 2.0 · 商用许可

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.

发布日期
2025-10-14
参数规模
4.0B
上下文长度
—
支持模态
image, text

能力雷达图

23
general
29
coding
37
reasoning
27
science
60
agents
90
multimodal

排行榜排名

领域#排名分数来源
代码能力榜543
16.0
AA
通用能力榜530
26.0
AA
多模态榜139
26.0
LS
科学能力587
18.0
AA

基准测试分数 (LLM Stats)

(LLM Stats (zeroeval))

Chat

MM-MT-Bench7.50 / 100自报
IFEvalGoogle Research (2023)82.3%自报

Factuality

SimpleQA48.0%自报

General

MMLU77.2%自报
MLVU-M75.3%自报
BFCL-v363.3%自报
Include61.4%自报

Language

MMLU-Redux81.5%自报
MMLU-Pro67.1%自报
MMLU-ProX59.4%自报

Math

MathVista-Mini73.7%自报
LiveBench 2024112560.9%自报
MathVision51.6%自报
AIME 202546.6%自报
HMMT2530.7%自报
PolyMATH28.8%自报

Multimodal

VideoMMMU56.2%自报
OSWorld26.2%自报

Reasoning

CharXiv-D76.2%自报
SuperGPQA40.3%自报
CharXiv-R39.7%自报
LiveCodeBench v637.9%自报

Video

CharadesSTA55.5%自报

Vision

DocVQAtest95.3%自报
ScreenSpot94.0%自报
OCRBench88.1%自报
MMBench-V1.185.1%自报
AI2D84.1%自报
InfoVQAtest80.3%自报
CC-OCR76.2%自报
RealWorldQA70.9%自报
MMStar69.8%自报
MVBench68.9%自报
MMMU (val)67.4%自报
BLINK65.8%自报
MuirBench63.8%自报
OCRBench-V2 (en)63.7%自报
ScreenSpot Pro59.5%自报
OCRBench-V2 (zh)57.6%自报
Hallusion Bench57.6%自报
LVBench56.2%自报
MMMU-Pro53.2%自报
ODinW48.2%自报
ERQA41.3%自报

Writing

WritingBench82.5%自报

AA 评测指数

(Artificial Analysis)
Mmlu Pro(TIGER-Lab (Univ. of Waterloo, Toronto, CMU, 2024))
63.4
Gpqa(NYU + Cohere + Anthropic (2023))
37.1
Math Index(Artificial Analysis)
37.0
Aime 25(MAA (Mathematical Association of America))
37.0
Ifbench(Google Research (2023))
31.8
Livecodebench(UC Berkeley + MIT + Cornell (2024))
29.0
Tau2(Sierra + U Toronto + Vector Institute (2025))
23.4
Lcr(Artificial Analysis)
14.0
Intelligence Index(Artificial Analysis)
5.7
Hle(Center for AI Safety + Scale AI (2025))
3.6
Terminalbench Hard(Stanford × Laude Institute (2026))
0.0

LLM Stats 分类评分

(LLM Stats (zeroeval))
Chat
4
Communication
4
Multimodal
90
Instruction Following
80
General
80
Grounding
80
Creativity
80
Text-to-image
80
Writing
80
Image To Text
70
Language
70
Legal
70
Structured Output
70
3d
70
Long Context
60
Math
60
Reasoning
60
Spatial Reasoning
60
Finance
60
Healthcare
60
Tool Calling
60
Video
60
Vision
60
Factuality
50
Physics
40
Agents
40
Chemistry
40
Economics
40

定价

输入价格免费
输出价格免费
混合价格(3:1)免费

速度

Tokens/秒0.0
首Token延迟0.00s
首回答延迟0.00s

供应商价格排行

暂无提供商数据

外部链接