Перейти к основному содержанию
AITier

Qwen3.5 27B (Reasoning)

AlibabaQwenOpen WeightApache 2.0 · Commercial OK

Описание

Qwen3.5-27B is a multimodal dense foundation model with 27 billion parameters. It combines strong reasoning, coding, multilingual, long-context, and visual understanding performance in a production-friendly open-weight package with a native 262K context window.

Дата выхода
2026-02-24
Параметры
27.0B
Длина контекста
262K
Модальности
image, text, video

Радар способностей

38
general
36
coding
86
reasoning
57
scienceоцен.
60
agents
80
multimodal

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

Рейтинги

Домен#МестоОценкаИсточник
Agents & Tools51
57.0
LS
Code Ranking84
65.0
AA
General Ranking43
80.0
AA
Multimodal Ranking60
70.0
LS
Reasoning54
67.0
LS
Science63
68.0
AA

Оценки бенчмарков (LLM Stats)

3d

SUNRGBD0.35 / 100Сам.
Hypersim0.13 / 100Сам.

Agents

t2-bench79.0%Сам.
BFCL-V468.5%Сам.
AndroidWorld_SR64.2%Сам.
WideSearch61.1%Сам.
BrowseComp61.0%Сам.
FullStackBench en60.1%Сам.
TIR-Bench59.8%Сам.
FullStackBench zh57.4%Сам.
OSWorld-Verified56.2%Сам.
VITA-Bench41.9%Сам.
Terminal-Bench 2.041.6%Сам.
DeepPlanning22.6%Сам.

Biology

GPQA85.5%Сам.

Chemistry

SuperGPQA65.6%Сам.

Code

SWE-Bench Verified72.4%Сам.

Communication

Multi-Challenge60.8%Сам.

Embodied

EmbSpatialBench0.84 / 100Сам.

Finance

MMLU-Pro86.1%Сам.
MMLU-ProX82.2%Сам.

General

IFEval95.0%Сам.
MMLU-Redux93.2%Сам.
C-Eval90.5%Сам.
MAXIFE88.0%Сам.
Global PIQA87.5%Сам.
MMMLU85.9%Сам.
MMMU82.3%Сам.
Include81.6%Сам.
MMStar81.0%Сам.
LiveCodeBench v680.7%Сам.
IFBench76.5%Сам.
MMMU-Pro75.0%Сам.
LongBench v260.6%Сам.
NOVA-6358.1%Сам.
SimpleVQA0.56 / 100Сам.

Grounding

RefCOCO-avg0.91 / 100Сам.
ScreenSpot Pro70.3%Сам.
RefSpatialBench0.68 / 100Сам.

Healthcare

VideoMMMU82.3%Сам.
SlakeVQA80.0%Сам.
MedXpertQA62.4%Сам.
PMC-VQA62.4%Сам.

Image To Text

OCRBench89.4%Сам.

Language

LingoQA82.0%Сам.
WMT24++77.6%Сам.

Long Context

MLVU85.9%Сам.
LVBench73.6%Сам.
AA-LCR66.1%Сам.
MMLongBench-Doc0.60 / 100Сам.

Math

HMMT 202592.0%Сам.
HMMT2589.8%Сам.
MathVista-Mini87.8%Сам.
DynaMath87.7%Сам.
MathVision86.0%Сам.
CodeForces0.81 / 3000Сам.
PolyMATH71.2%Сам.
Humanity's Last Exam48.5%Сам.

Multimodal

VLMsAreBlind96.9%Сам.
V*93.7%Сам.
AI2D92.9%Сам.
MMBench-V1.192.6%Сам.
OmniDocBench 1.588.9%Сам.
VideoMME w sub.87.0%Сам.
VideoMME w/o sub.82.8%Сам.
CC-OCR81.0%Сам.
CharXiv-R79.5%Сам.
MVBench74.6%Сам.
MMVU73.3%Сам.
BabyVision44.6%Сам.
ZEROBench-Sub0.36 / 100Сам.
Nuscene15.2%Сам.
ZEROBench0.10 / 100Сам.

Reasoning

CountBench0.98 / 100Сам.
Hallusion Bench70.0%Сам.
BrowseComp-zh62.1%Сам.
ERQA60.5%Сам.
Seal-047.2%Сам.
OJBench40.1%Сам.

Spatial Reasoning

RealWorldQA83.7%Сам.

Vision

ODinW41.1%Сам.

Индексы оценки AA

Intelligence Index
42.1
Coding Index
34.9
Tau2
0.9
Gpqa
0.9
Ifbench
0.8
Lcr
0.7
Scicode
0.4
Terminalbench Hard
0.3
Hle
0.2

Оценки категорий LLM Stats

Biology
90
Instruction Following
90
Structured Output
80
Text-to-image
80
Video
80
Chemistry
80
Embodied
80
Finance
80
General
80
Grounding
80
Image To Text
80
Language
80
Legal
80
Math
80
Physics
80
Spatial Reasoning
70
Vision
70
Economics
70
Frontend Development
70
Healthcare
70
Long Context
70
Multimodal
70
Reasoning
70
Tool Calling
60
Agents
60
Code
60
Communication
60
Search
60
Spatial
20
3d
20

Цены

Цена ввода$0.3 / 1M tokens
Цена вывода$2.4 / 1M tokens
Смешанная цена (3:1)$0.825 / 1M tokens

Скорость

Токенов/сек87.6 tokens/s
Задержка первого токена1.40s
Время до первого ответа24.23s

Доступные провайдеры

(Внутренние единицы LS)
ПровайдерЦена вводаЦена вывода
Novita300K2.4M

Внешние ссылки

LLM Stats