跳转到主要内容
AITier

Qwen3.5 27B (Reasoning)

AlibabaQwen开源权重Apache 2.0 · 商用许可

描述

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.

发布日期
2026-02-24
参数规模
27.0B
上下文长度
262K
支持模态
audio, image, text, video

能力雷达图

32
general
40
coding
86
reasoning
57
science
60
agents
80
multimodal

排行榜排名

领域#排名分数来源
智能体能力模型榜105
38.0
LS
代码能力榜131
68.0
AA
通用能力榜92
73.0
AA
多模态榜36
54.0
LS
科学能力122
66.0
AA

基准测试分数 (LLM Stats)

(LLM Stats (zeroeval))

3d

SUNRGBD0.35 / 100自报
Hypersim0.13 / 100自报

Agents

t2-bench79.0%自报
BFCL-V468.5%自报
AndroidWorld_SR64.2%自报
WideSearch61.1%自报
BrowseCompOpenAI (2025)61.0%自报
FullStackBench en60.1%自报
TIR-Bench59.8%自报
FullStackBench zh57.4%自报
OSWorld-Verified56.2%自报
VITA-Bench41.9%自报
Terminal-Bench 2.0Stanford × Laude Institute (2026)41.6%自报
DeepPlanning22.6%自报

Biology

GPQANYU + Cohere + Anthropic (2023)85.5%自报

Chemistry

SuperGPQA65.6%自报

Code

SWE-Bench Verified72.4%自报

Communication

Multi-Challenge60.8%自报

Embodied

EmbSpatialBench0.84 / 100自报

Finance

MMLU-Pro86.1%自报
MMLU-ProX82.2%自报

General

IFEvalGoogle Research (2023)95.0%自报
MMLU-Redux93.2%自报
C-Eval90.5%自报
MAXIFE88.0%自报
Global PIQA87.5%自报
MMMLU85.9%自报
MMMU82.3%自报
Include81.6%自报
MMStar81.0%自报
LiveCodeBench v680.7%自报
IFBench76.5%自报
MMMU-Pro75.0%自报
LongBench v260.6%自报
NOVA-6358.1%自报
SimpleVQA0.56 / 100自报

Grounding

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

Healthcare

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

Image To Text

OCRBench89.4%自报

Language

LingoQA82.0%自报
WMT24++77.6%自报

Long Context

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

Math

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

Multimodal

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

Reasoning

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

Spatial Reasoning

RealWorldQA83.7%自报

Vision

ODinW41.1%自报

AA 评测指数

(Artificial Analysis)
Intelligence Index(Artificial Analysis)
34.6
Tau2(Sierra + U Toronto + Vector Institute (2025))
0.9
Gpqa(NYU + Cohere + Anthropic (2023))
0.9
Ifbench(Google Research (2023))
0.8
Lcr(Artificial Analysis)
0.7
Scicode(UIUC + Argonne National Lab (2024))
0.4
Terminalbench Hard(Stanford × Laude Institute (2026))
0.3
Hle(Center for AI Safety + Scale AI (2025))
0.2

LLM Stats 分类评分

(LLM Stats (zeroeval))
Instruction Following
90
Biology
90
Legal
80
Math
80
Physics
80
Structured Output
80
Image To Text
80
Language
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

定价

输入价格$0.3 / 1M tokens
输出价格$2.4 / 1M tokens
混合价格(3:1)$0.825 / 1M tokens

速度

Tokens/秒0.0
首Token延迟0.00s
首回答延迟0.00s

供应商价格排行

供应商价格排行

4 个供应商

最便宜: Novita最贵: Alibaba
供应商输入输出
1Novita最便宜
$0
$0
2EmpirioLabs AI
$0.086
$0.688
3NanoGPT
$0.27
$2.16
4Alibaba主要
$0.3
$2.4

比较该模型在不同 API 供应商之间的定价。

外部链接

LLM StatsArtificial Analysis