Skip to main content

Qwen3 VL 4B Instruct

AlibabaQwenOpen WeightApache 2.0 · Commercial OK

Description

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.

Release Date
2025-10-14
Parameters
4.0B
Context Length
Modalities
image, text

Capability Radar

22
general
26
coding
37
reasoning
22
science
60
agents
90
multimodal

Rankings

Domain#RankScoreSource
Agentic Capability141
29.0
LS
Code Ranking464
16.0
AA
General Ranking468
26.0
AA
Multimodal Ranking82
27.0
LS
Science508
20.0
AA

Benchmark Scores (LLM Stats)

(LLM Stats (zeroeval))

3d

BLINK65.8%SR

Agents

BFCL-v363.3%SR
OSWorld26.2%SR

Chemistry

SuperGPQA40.3%SR

Communication

MM-MT-Bench7.50 / 100SR
WritingBench82.5%SR

Factuality

SimpleQA48.0%SR

Finance

MMLU77.2%SR
MMLU-Pro67.1%SR
MMLU-ProX59.4%SR

General

IFEvalGoogle Research (2023)82.3%SR
MMLU-Redux81.5%SR
MLVU-M75.3%SR
MMStar69.8%SR
MMMU (val)67.4%SR
Include61.4%SR
LiveBench 2024112560.9%SR
MMMU-Pro53.2%SR
LiveCodeBench v637.9%SR

Grounding

ScreenSpot94.0%SR
ScreenSpot Pro59.5%SR

Healthcare

VideoMMMU56.2%SR

Image To Text

OCRBench88.1%SR
OCRBench-V2 (en)63.7%SR
OCRBench-V2 (zh)57.6%SR

Language

CharadesSTA55.5%SR

Long Context

LVBench56.2%SR

Math

MathVista-Mini73.7%SR
MathVision51.6%SR
AIME 202546.6%SR
HMMT2530.7%SR
PolyMATH28.8%SR

Multimodal

DocVQAtest95.3%SR
MMBench-V1.185.1%SR
AI2D84.1%SR
InfoVQAtest80.3%SR
CharXiv-D76.2%SR
CC-OCR76.2%SR
MVBench68.9%SR
MuirBench63.8%SR
CharXiv-R39.7%SR

Reasoning

Hallusion Bench57.6%SR
ERQA41.3%SR

Spatial Reasoning

RealWorldQA70.9%SR

Vision

ODinW48.2%SR

AA Evaluation Indices

(Artificial Analysis)
Math Index(Artificial Analysis)
37.0
Intelligence Index(Artificial Analysis)
3.7
Mmlu Pro(TIGER-Lab (Univ. of Waterloo, Toronto, CMU, 2024))
0.6
Gpqa(NYU + Cohere + Anthropic (2023))
0.4
Aime 25(MAA (Mathematical Association of America))
0.4
Ifbench(Google Research (2023))
0.3
Livecodebench(UC Berkeley + MIT + Cornell (2024))
0.3
Tau2(Sierra + U Toronto + Vector Institute (2025))
0.2
Lcr(Artificial Analysis)
0.1
Scicode(UIUC + Argonne National Lab (2024))
0.1
Hle(Center for AI Safety + Scale AI (2025))
0.0
Terminalbench Hard(Stanford × Laude Institute (2026))
0.0

LLM Stats Category Scores

(LLM Stats (zeroeval))
Communication
4
Multimodal
90
Instruction Following
80
General
80
Grounding
80
Creativity
80
Text-to-image
80
Writing
80
Legal
70
Structured Output
70
Image To Text
70
Language
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

Pricing

Input PriceFree
Output PriceFree
Blended Price (3:1)Free

Speed

Tokens/sec0.0
Time to First Token0.00s
Time to Answer0.00s

Provider Price Ranking

Provider Price Ranking

1 providers

ProviderInputOutput
1DeepInfra
$0
$0

Compare pricing across different API providers for this model.

External Sources