Qwen3 VL 4B Instruct

Name: Qwen3 VL 4B Instruct
Author: Alibaba

Multimodal

Alibaba

Qwen3 VL 4B Instruct is a compact 4B-parameter multimodal vision-language model from the Qwen team, released in September 2025. It provides strong document and GUI understanding for its size, scoring 0.95 on DocVQA and 0.94 on ScreenSpot. It is Apache 2.0 licensed and open-weight.

Key Specifications

Parameters

4.0B

Context

262.1K

Release Date

September 22, 2025

Average Score

61.7%

API Documentation Research Paper Results Blog

Timeline

Key dates in the model's history

Announcement

September 22, 2025

Last Update

August 27, 2026

Today

September 10, 2026

Technical Specifications

Parameters

4.0B

Training Tokens

Knowledge Cutoff

Family

Capabilities

MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)

$0.10

Output (per 1M tokens)

$0.60

Max Input Tokens

262.1K

Max Output Tokens

262.1K

Supported Features

Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

General Knowledge

Tests on general knowledge and understanding

MMLU

• Self-reported

77.2%

Multimodal

Working with images and visual data

AI2D

TEST • Self-reported

84.1%

Other Tests

Specialized benchmarks

DocVQA

Self-reported by the model provider • Self-reported

95.0%

ScreenSpot

Self-reported by the model provider • Self-reported

94.0%

OCRBench

Self-reported by the model provider • Self-reported

88.0%

AIME 2025

• Self-reported

46.6%

BFCL-v3

• Self-reported

63.3%

BLINK

• Self-reported

65.8%

CC-OCR

overall • Self-reported

76.2%

CharadesSTA

• Self-reported

55.5%

CharXiv-D

DQ • Self-reported

76.2%

CharXiv-R

RQ • Self-reported

39.7%

ERQA

• Self-reported

41.3%

Hallusion Bench

• Self-reported

57.6%

HMMT25

• Self-reported

30.7%

IFEval

• Self-reported

82.3%

Include

• Self-reported

61.4%

InfoVQAtest

• Self-reported

80.3%

LiveBench 20241125

• Self-reported

60.9%

LiveCodeBench v6

25.02-25.05 • Self-reported

37.9%

LVBench

• Self-reported

56.2%

MathVision

• Self-reported

51.6%

MathVista-Mini

• Self-reported

73.7%

MLVU-M

MCQ • Self-reported

75.3%

MM-MT-Bench

• Self-reported

7.5%

MMBench-V1.1

en, V1.1 • Self-reported

85.1%

MMLU-Pro

• Self-reported

67.1%

MMLU-ProX

• Self-reported

59.4%

MMLU-Redux

• Self-reported

81.5%

MMMU (val)

AI • Self-reported

67.4%

MMMU-Pro

full • Self-reported

53.2%

MMStar

• Self-reported

69.8%

MuirBench

• Self-reported

63.8%

MVBench

• Self-reported

68.9%

OCRBench-V2 (en)

en/zh - using en value • Self-reported

63.7%

OCRBench-V2 (zh)

zh • Self-reported

57.6%

ODinW

13 • Self-reported

48.2%

OSWorld

• Self-reported

26.2%

PolyMATH

• Self-reported

28.8%

RealWorldQA

• Self-reported

70.9%

ScreenSpot Pro

• Self-reported

59.5%

SimpleQA

VQA • Self-reported

48.0%

SuperGPQA

• Self-reported

40.3%

VideoMMMU

• Self-reported

56.2%

WritingBench

• Self-reported

82.5%

License & Metadata

License

apache-2.0

Announcement Date

September 22, 2025

Last Updated

August 27, 2026

Similar Models

All Models

Qwen3.5-4B

Alibaba

MM4.0B

Best score:0.8 (GPQA)

Released:Mar 2026

Qwen3 VL 4B Thinking

Alibaba

MM4.0B

Best score:0.8 (MMLU)

Released:Sep 2025

Price:$0.10/1M tokens

Qwen2.5 VL 7B Instruct

Alibaba

MM8.3B

Released:Jan 2025

Qwen2.5-Omni-7B

Alibaba

MM7.0B

Best score:0.8 (HumanEval)

Released:Mar 2025

Qwen2.5 VL 32B Instruct

Alibaba

MM33.5B

Best score:0.9 (HumanEval)

Released:Feb 2025

Qwen3 VL 32B Thinking

Alibaba

MM33.0B

Best score:0.9 (MMLU)

Released:Sep 2025

Qwen3.5 9B

Alibaba

9.0B

Best score:0.8 (GPQA)

Released:Mar 2026

Qwen2.5-Coder 7B Instruct

Alibaba

7.0B

Best score:0.9 (HumanEval)

Released:Sep 2024

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.