Passer au contenu principal
AITier

Qwen3.5 27B (Reasoning)

AlibabaQwenOpen WeightApache 2.0 · Usage Commercial

Description

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.

Date de sortie
2026-02-24
Paramètres
27.0B
Longueur du contexte
262K
Modalités
audio, image, text, video

Radar de capacités

23
general
60
coding
86
reasoning
67
science
60
agents
80
multimodal

Classements

Domaine#RangScoreSource
Capacité agentique57
48.0
LS
Classement codage189
69.0
AA
Classement général122
64.0
AA
Classement multimodal74
53.0
LS
Science169
63.0
AA

Scores de benchmarks (LLM Stats)

(LLM Stats (zeroeval))

Agents

t2-bench79.0%Aut.
VITA-Bench41.9%Aut.
DeepPlanning22.6%Aut.

Chat

IFEvalGoogle Research (2023)95.0%Aut.
Multi-Challenge60.8%Aut.

Code

FullStackBench en60.1%Aut.
FullStackBench zh57.4%Aut.

General

C-Eval90.5%Aut.
MAXIFE88.0%Aut.
Include81.6%Aut.
AndroidWorld_SR64.2%Aut.
NOVA-6358.1%Aut.

Healthcare

MedXpertQA62.4%Aut.
PMC-VQA62.4%Aut.

Instruction Following

IFBench76.5%Aut.

Language

MMLU-Redux93.2%Aut.
MMLU-Pro86.1%Aut.
MMMLU85.9%Aut.
MMLU-ProX82.2%Aut.
WMT24++77.6%Aut.

Long Context

LongBench v260.6%Aut.

Math

HMMT 202592.0%Aut.
HMMT2589.8%Aut.
MathVista-Mini87.8%Aut.
DynaMath87.7%Aut.
MathVision86.0%Aut.
CodeForces0.81 / 3000Aut.
PolyMATH71.2%Aut.

Multimodal

VideoMME w/o sub.82.8%Aut.
VideoMMMU82.3%Aut.
MMMU82.3%Aut.
TIR-Bench59.8%Aut.
OSWorld-Verified56.2%Aut.

Reasoning

Global PIQA87.5%Aut.
GPQANYU + Cohere + Anthropic (2023)85.5%Aut.
LiveCodeBench v680.7%Aut.
CharXiv-R79.5%Aut.
SWE-Bench Verified72.4%Aut.
AA-LCR66.1%Aut.
SuperGPQA65.6%Aut.
BrowseComp-zh62.1%Aut.
BrowseCompOpenAI (2025)61.0%Aut.
Humanity's Last Exam48.5%Aut.
Seal-047.2%Aut.
Terminal-Bench 2.0Stanford × Laude Institute (2026)41.6%Aut.
OJBench40.1%Aut.

Search

WideSearch61.1%Aut.

Tool Calling

BFCL-V468.5%Aut.

Video

MLVU85.9%Aut.

Vision

CountBench0.98 / 100Aut.
VLMsAreBlind96.9%Aut.
V*93.7%Aut.
AI2D92.9%Aut.
MMBench-V1.192.6%Aut.
RefCOCO-avg0.91 / 100Aut.
OCRBench89.4%Aut.
OmniDocBench 1.588.9%Aut.
VideoMME w sub.87.0%Aut.
EmbSpatialBench0.84 / 100Aut.
RealWorldQA83.7%Aut.
LingoQA82.0%Aut.
CC-OCR81.0%Aut.
MMStar81.0%Aut.
SlakeVQA80.0%Aut.
MMMU-Pro75.0%Aut.
MVBench74.6%Aut.
LVBench73.6%Aut.
MMVU73.3%Aut.
ScreenSpot Pro70.3%Aut.
Hallusion Bench70.0%Aut.
RefSpatialBench0.68 / 100Aut.
ERQA60.5%Aut.
MMLongBench-Doc0.60 / 100Aut.
SimpleVQA0.56 / 100Aut.
BabyVision44.6%Aut.
ODinW41.1%Aut.
ZEROBench-Sub0.36 / 100Aut.
SUNRGBD0.35 / 100Aut.
Nuscene15.2%Aut.
Hypersim0.13 / 100Aut.
ZEROBench0.10 / 100Aut.

Indices d'évaluation AA

(Artificial Analysis)
Tau2(Sierra + U Toronto + Vector Institute (2025))
93.9
Gpqa(NYU + Cohere + Anthropic (2023))
85.8
Lcr(Artificial Analysis)
77.7
Ifbench(Google Research (2023))
75.6
Terminalbench Hard(Stanford × Laude Institute (2026))
32.6
Hle(Center for AI Safety + Scale AI (2025))
23.9
Intelligence Index(Artificial Analysis)
22.9

Scores par catégorie LLM Stats

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

Tarification

Prix d'entrée$0.3 / 1M tokens
Prix de sortie$2.4 / 1M tokens
Prix mixte (3:1)$0.825 / 1M tokens

Vitesse

Tokens/sec0.0
Délai du premier token0.00s
Temps de réponse0.00s

Classement des Prix par Fournisseur

Classement des Prix par Fournisseur

4 fournisseurs

Moins cher: DeepInfraPlus cher: Alibaba
FournisseurEntréeSortie
1DeepInfraMoins cher
$0
$0
2Novita
$0
$0
3EmpirioLabs AI
$0.086
$0.688
4AlibabaPRINCIPAL
$0.3
$2.4

Comparer les prix entre différents fournisseurs API pour ce modèle.

Sources externes

LLM StatsArtificial Analysis