Перейти к основному содержанию

Qwen3 VL 30B A3B (Reasoning)

AlibabaQwenОткрытые весаApache 2.0 · Коммерческое использование

Описание

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.

Дата выхода
2025-10-03
Параметры
31.0B
Длина контекста
131K
Модальности
image, text, video

Радар способностей

33
general
61
coding
81
reasoning
44
scienceоцен.
70
agents
100
multimodal

Science использует прокси на основе рассуждений, когда специализированные научные бенчмарки недоступны.

Рейтинги

Оценки бенчмарков (LLM Stats)

3d

BLINK65.4%Сам.

Agents

BFCL-v368.6%Сам.
OSWorld30.6%Сам.

Biology

GPQA74.4%Сам.

Chemistry

SuperGPQA56.4%Сам.

Communication

MM-MT-Bench7.90 / 100Сам.
WritingBench85.2%Сам.
Multi-IF73.0%Сам.

Creativity

Creative Writing v382.5%Сам.
Arena-Hard v256.7%Сам.

Factuality

SimpleQA23.9%Сам.

Finance

MMLU87.6%Сам.
MMLU-Pro80.5%Сам.
MMLU-ProX76.1%Сам.

General

MMLU-Redux90.9%Сам.
IFEval81.7%Сам.
MLVU-M78.9%Сам.
MMMU (val)76.0%Сам.
MMStar75.5%Сам.
Include74.5%Сам.
LiveBench 2024112572.1%Сам.
LiveCodeBench v664.2%Сам.
MMMU-Pro63.0%Сам.

Grounding

ScreenSpot94.7%Сам.
ScreenSpot Pro57.3%Сам.

Healthcare

VideoMMMU75.0%Сам.

Image To Text

OCRBench83.9%Сам.
OCRBench-V2 (en)62.6%Сам.
OCRBench-V2 (zh)60.4%Сам.

Language

CharadesSTA62.7%Сам.

Long Context

LVBench59.2%Сам.

Math

AIME 202583.1%Сам.
MathVista-Mini81.9%Сам.
HMMT2567.6%Сам.
MathVision65.7%Сам.
PolyMATH51.7%Сам.

Multimodal

DocVQAtest95.0%Сам.
MMBench-V1.188.9%Сам.
CharXiv-D86.9%Сам.
AI2D86.9%Сам.
InfoVQAtest86.0%Сам.
CC-OCR77.8%Сам.
MuirBench77.6%Сам.
Video-MME73.3%Сам.
MVBench72.0%Сам.
CharXiv-R56.6%Сам.

Reasoning

Hallusion Bench66.0%Сам.
ERQA45.3%Сам.

Spatial Reasoning

RealWorldQA77.4%Сам.

Vision

ODinW42.3%Сам.

Индексы оценки AA

Math Index
82.3
Intelligence Index
13.3
Aime 25
0.8
Mmlu Pro
0.8
Gpqa
0.7
Livecodebench
0.7
Ifbench
0.5
Lcr
0.4
Scicode
0.3
Tau2
0.2
Hle
0.1
Terminalbench Hard
0.1

Оценки категорий LLM Stats

Communication
3
Multimodal
100
Instruction Following
80
Language
80
Legal
80
Structured Output
80
Finance
80
Grounding
80
Healthcare
80
Text-to-image
80
Image To Text
70
Math
70
Physics
70
Reasoning
70
Spatial Reasoning
70
General
70
3d
70
Biology
70
Chemistry
70
Creativity
70
Tool Calling
70
Video
70
Vision
70
Writing
70
Long Context
60
Economics
60
Agents
50
Factuality
20

Цены

Цена ввода$0.2 / 1M токенов
Цена вывода$0.75 / 1M токенов
Смешанная цена (3:1)$0.338 / 1M токенов

Скорость

Токенов/сек122.7
Задержка первого токена1.01s
Время до первого ответа17.31s

Рейтинг цен провайдеров

Рейтинг цен провайдеров

2 провайдеров

Самый дешевый: Alibaba (China)Самый дорогой: Alibaba
ПровайдерВводВывод
1Alibaba (China)Самый дешевый
$0.108
$0.431
2AlibabaОсновной
$0.2
$0.75

Сравнение цен разных API-провайдеров для этой модели.

Внешние ссылки