跳轉到主要內容

Qwen3 VL 30B A3B Instruct

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
2025-10-03
參數規模
31.0B
上下文長度
262K
支援模態
image, text, video

能力雷達圖

28
general
48
coding
72
reasoning
50
science
70
agents
100
multimodal

排行榜排名

領域#排名分數來源
程式碼能力榜427
31.0
AA
通用能力榜464
30.0
AA
多模態榜120
38.0
LS
科學能力384
39.0
AA

基準測試分數 (LLM Stats)

(LLM Stats (zeroeval))

Chat

MM-MT-Bench8.10 / 100自報
IFEvalGoogle Research (2023)85.8%自報
Multi-IF66.1%自報

Creativity

Creative Writing v384.6%自報

Factuality

SimpleQA27.0%自報

General

MMLU85.0%自報
MLVU-M81.3%自報
Include71.6%自報
BFCL-v366.3%自報
Arena-Hard v258.5%自報

Language

MMLU-Redux88.4%自報
MMLU-Pro77.8%自報
MMLU-ProX70.9%自報

Math

MathVista-Mini80.1%自報
AIME 202569.3%自報
LiveBench 2024112565.4%自報
MathVision60.2%自報
HMMT2550.6%自報
PolyMATH44.3%自報

Multimodal

Video-MME74.5%自報
VideoMMMU68.7%自報
OSWorld30.3%自報

Reasoning

CharXiv-D85.5%自報
GPQANYU + Cohere + Anthropic (2023)70.4%自報
SuperGPQA53.1%自報
CharXiv-R48.9%自報
LiveCodeBench v642.6%自報

Video

CharadesSTA63.5%自報

Vision

DocVQAtest95.0%自報
ScreenSpot94.7%自報
OCRBench90.3%自報
MMBench-V1.187.0%自報
AI2D85.0%自報
InfoVQAtest82.0%自報
CC-OCR80.7%自報
MMMU (val)74.2%自報
RealWorldQA73.7%自報
MVBench72.3%自報
MMStar72.1%自報
BLINK67.7%自報
OCRBench-V2 (en)63.2%自報
MuirBench62.9%自報
LVBench62.5%自報
Hallusion Bench61.5%自報
ScreenSpot Pro60.5%自報
MMMU-Pro60.4%自報
OCRBench-V2 (zh)57.8%自報
ODinW47.5%自報
ERQA43.0%自報

Writing

WritingBench82.6%自報

AA 評測指數

(Artificial Analysis)
Mmlu Pro(TIGER-Lab (Univ. of Waterloo, Toronto, CMU, 2024))
76.4
Aime 25(MAA (Mathematical Association of America))
72.3
Math Index(Artificial Analysis)
72.3
Gpqa(NYU + Cohere + Anthropic (2023))
69.5
Livecodebench(UC Berkeley + MIT + Cornell (2024))
47.6
Ifbench(Google Research (2023))
33.1
Tau2(Sierra + U Toronto + Vector Institute (2025))
19.0
Intelligence Index(Artificial Analysis)
7.9
Hle(Center for AI Safety + Scale AI (2025))
6.3
Terminalbench Hard(Stanford × Laude Institute (2026))
6.1

LLM Stats 分類評分

(LLM Stats (zeroeval))
Chat
3
Communication
3
Multimodal
100
Instruction Following
80
Language
80
Structured Output
80
General
80
Grounding
80
Creativity
80
Text-to-image
80
Writing
80
Image To Text
70
Legal
70
Math
70
Reasoning
70
Spatial Reasoning
70
Finance
70
Healthcare
70
3d
70
Biology
70
Tool Calling
70
Video
70
Vision
70
Long Context
60
Physics
60
Chemistry
60
Agents
50
Economics
50
Factuality
30

定價

輸入價格$0.2 / 1M tokens
輸出價格$0.8 / 1M tokens
混合價格(3:1)$0.35 / 1M tokens

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

供應商價格排行

6 個供應商

最便宜: DeepInfra最貴: NovitaAI
供應商輸入輸出
1DeepInfra最便宜
$0
$0
2Kilo Gateway
$0.13
$0.52
3OpenRouter
$0.15
$0.6
4DevPass (LLM Gateway)
$0.15
$0.6
5Alibaba主要
$0.2
$0.8
6NovitaAI
$0.2
$0.7

比較該模型在不同 API 供應商之間的定價。

外部連結