Перейти к основному содержанию
AITier

Qwen3.6 Plus

AlibabaQwenProprietary

Описание

Qwen3.6 Plus is Alibaba's next-generation flagship model featuring a 1 million token native context window, up to 65,536 output tokens, and always-on chain-of-thought reasoning. It uses a next-generation hybrid architecture optimized for efficiency and scalability. It leads on Terminal-Bench 2.0 agentic coding (61.6), surpassing Claude 4.5 Opus, and achieves strong results on document understanding (OmniDocBench 91.2) and multimodal reasoning (MMMU 86.0). Compared to Qwen 3.5, it is significantly more decisive in reasoning, using fewer tokens on straightforward tasks with better agent stability.

Дата выхода
2026-04-02
Параметры
—
Длина контекста
1.0M
Модальности
image, text, video

Радар способностей

37
general
52
coding
88
reasoning
59
scienceоцен.
60
agents
90
multimodal

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

Рейтинги

Домен#МестоОценкаИсточник
Агентные возможности68
54.0
LS
Рейтинг кодинга52
76.0
AA
Общий рейтинг23
80.0
AA
Мультимодальный рейтинг17
87.0
LS
Рассуждения29
82.0
LS
Наука62
69.0
AA

Оценки бенчмарков (LLM Stats)

Agents

GDPval-AA1160.00 / 3000Сам.
WideSearch74.3%Сам.
MCP Atlas74.1%Сам.
TAU3-Bench70.7%Сам.
OSWorld-Verified62.5%Сам.
TIR-Bench61.6%Сам.
Terminal-Bench 2.061.6%Сам.
Claw-Eval58.7%Сам.
SWE-Bench Pro56.6%Сам.
MCP-Mark48.2%Сам.
SkillsBench45.7%Сам.
VITA-Bench44.3%Сам.
DeepPlanning41.5%Сам.
Finance Agent v240.8%Сам.
Toolathlon39.8%Сам.
NL2Repo37.9%Сам.
FrontierSWE22.0%Сам.

Biology

GPQA90.4%Сам.

Chemistry

SuperGPQA71.6%Сам.

Code

SWE-Bench Verified78.8%Сам.
SWE-bench Multilingual73.8%Сам.

Finance

MMLU-Pro88.5%Сам.
MMLU-ProX84.7%Сам.

General

MMLU-Redux94.5%Сам.
IFEval94.3%Сам.
C-Eval93.3%Сам.
Global PIQA89.8%Сам.
MMMLU89.5%Сам.
MAXIFE88.2%Сам.
LiveCodeBench v687.1%Сам.
MMMU86.0%Сам.
Include85.1%Сам.
MMStar83.3%Сам.
MMMU-Pro78.8%Сам.
IFBench74.2%Сам.
LiveBench70.9%Сам.
SimpleVQA0.67 / 100Сам.
LongBench v262.0%Сам.
NOVA-6357.9%Сам.

Grounding

RefCOCO-avg0.94 / 100Сам.
ScreenSpot Pro68.2%Сам.

Healthcare

VideoMMMU84.0%Сам.

Language

WMT24++84.3%Сам.

Long Context

MLVU86.7%Сам.
AA-LCR68.3%Сам.
MMLongBench-Doc0.62 / 100Сам.

Math

HMMT 202596.7%Сам.
AIME 202695.3%Сам.
HMMT2594.6%Сам.
We-Math89.0%Сам.
DynaMath88.0%Сам.
MathVision88.0%Сам.
HMMT Feb 2687.8%Сам.
IMO-AnswerBench83.8%Сам.
PolyMATH77.4%Сам.
Humanity's Last Exam28.8%Сам.

Multimodal

V*96.9%Сам.
AI2D94.4%Сам.
OmniDocBench 1.591.2%Сам.
Video-MME84.2%Сам.
CC-OCR83.4%Сам.
CharXiv-R81.5%Сам.

Reasoning

CountBench0.98 / 100Сам.
ERQA65.7%Сам.

Spatial Reasoning

RealWorldQA85.4%Сам.

Vision

ODinW51.8%Сам.

Индексы оценки AA

Coding Index
54.5
Intelligence Index
39.6
Tau2
1.0
Gpqa
0.9
Ifbench
0.8
Lcr
0.7
Terminalbench V2 1
0.6
Terminalbench Hard
0.4
Scicode
0.4
Hle
0.3
Tau Banking
0.2

Оценки категорий LLM Stats

Legal
100
Finance
100
Agents
69
General
54
Reasoning
28
Language
90
Biology
90
Video
90
Instruction Following
80
Math
80
Multimodal
80
Physics
80
Spatial Reasoning
80
Structured Output
80
Frontend Development
80
Grounding
80
Healthcare
80
Chemistry
80
Text-to-image
80
Vision
80
Image To Text
70
Long Context
70
Search
70
Economics
70
Code
60
Tool Calling
60
Coding
50

Цены

Цена ввода$0.5 / 1M токенов
Цена вывода$3 / 1M токенов
Смешанная цена (3:1)$1.125 / 1M токенов
Цена чтения кэша$0.05 / 1M токенов
Цена записи кэша$0.625 / 1M токенов

Скорость

Токенов/сек52.3
Задержка первого токена1.50s
Время до первого ответа107.64s

Рейтинг цен провайдеров

Рейтинг цен провайдеров

17 провайдеров

Самый дешевый: TogetherСамый дорогой: Venice AI
ПровайдерВводВывод
1TogetherСамый дешевый
$0
$0
2AIHubMix
$0.28
$1.69
3OpenRouter
$0.325
$1.95
4Kilo Gateway
$0.325
$1.95
5NanoGPT
$0.45
$2.7
6AlibabaОсновной
$0.5
$3
7OpenCode Go
$0.5
$3
8Alibaba (China)
$0.5
$3
9ZenMux
$0.5
$3
10FrogBot
$0.5
$3
11Vercel AI Gateway
$0.5
$3
12LLM Gateway
$0.5
$3
13Together AI
$0.5
$3
14Auriko
$0.5
$3
15OrcaRouter
$0.5
$3
16Merge Gateway
$0.5
$3
17Venice AI
$0.625
$3.75

Сравнение цен разных API-провайдеров для этой модели.

Внешние ссылки

LLM StatsArtificial Analysis