跳转到主要内容

Qwen3 VL 4B (Reasoning)

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

26
general
29
coding
30
reasoning
29
science
70
agents
100
multimodal

排行榜排名

领域#排名分数来源
智能体能力模型榜110
37.0
LS
代码能力榜412
22.0
AA
通用能力榜429
30.0
AA
多模态榜85
22.0
LS
科学能力464
27.0
AA

基准测试分数 (LLM Stats)

(LLM Stats (zeroeval))

3d

BLINK63.4%自报

Agents

BFCL-v367.3%自报
OSWorld31.4%自报

Biology

GPQANYU + Cohere + Anthropic (2023)64.1%自报

Chemistry

SuperGPQA46.8%自报

Communication

MM-MT-Bench7.70 / 100自报
WritingBench84.0%自报
Multi-IF73.6%自报

Creativity

Creative Writing v376.1%自报
Arena-Hard v236.8%自报

Finance

MMLU81.5%自报
MMLU-Pro73.6%自报
MMLU-ProX65.0%自报

General

MMLU-Redux86.0%自报
IFEvalGoogle Research (2023)82.6%自报
MLVU-M75.7%自报
MMStar73.2%自报
MMMU (val)70.8%自报
LiveBench 2024112568.4%自报
Include64.6%自报
MMMU-Pro57.0%自报
LiveCodeBench v651.3%自报

Grounding

ScreenSpot92.9%自报
ScreenSpot Pro49.2%自报

Healthcare

VideoMMMU69.4%自报

Image To Text

OCRBench80.8%自报
OCRBench-V2 (en)61.8%自报
OCRBench-V2 (zh)55.8%自报

Language

CharadesSTA59.0%自报

Long Context

LVBench53.5%自报

Math

MathVista-Mini79.5%自报
AIME 202574.5%自报
MathVision60.0%自报
HMMT2553.1%自报
PolyMATH44.6%自报

Multimodal

DocVQAtest94.2%自报
MMBench-V1.186.7%自报
AI2D84.9%自报
CharXiv-D83.9%自报
InfoVQAtest83.0%自报
MuirBench75.0%自报
CC-OCR73.8%自报
MVBench69.3%自报
CharXiv-R50.3%自报

Reasoning

Hallusion Bench64.1%自报
ERQA47.3%自报

Spatial Reasoning

RealWorldQA73.2%自报

Vision

ODinW39.4%自报

AA 评测指数

(Artificial Analysis)
Math Index(Artificial Analysis)
25.7
Intelligence Index(Artificial Analysis)
7.7
Mmlu Pro(TIGER-Lab (Univ. of Waterloo, Toronto, CMU, 2024))
0.7
Gpqa(NYU + Cohere + Anthropic (2023))
0.5
Ifbench(Google Research (2023))
0.4
Livecodebench(UC Berkeley + MIT + Cornell (2024))
0.3
Aime 25(MAA (Mathematical Association of America))
0.3
Lcr(Artificial Analysis)
0.2
Scicode(UIUC + Argonne National Lab (2024))
0.2
Tau2(Sierra + U Toronto + Vector Institute (2025))
0.2
Hle(Center for AI Safety + Scale AI (2025))
0.0
Terminalbench Hard(Stanford × Laude Institute (2026))
0.0

LLM Stats 分类评分

(LLM Stats (zeroeval))
Communication
3
Multimodal
100
Structured Output
80
Instruction Following
80
General
80
Legal
70
Math
70
Reasoning
70
Image To Text
70
Language
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

供应商价格排行

1 个供应商

供应商输入输出
1DeepInfra
$0
$0

比较该模型在不同 API 供应商之间的定价。

外部链接