跳转到主要内容
AITier

Qwen3.5 35B A3B (Non-reasoning)

AlibabaQwen开源权重Apache 2.0 · 商用许可

描述

Qwen3.5-35B-A3B is a multimodal Mixture-of-Experts model with 35 billion total parameters and 3 billion activated parameters. It combines strong reasoning, coding, agentic, and visual understanding performance with production-friendly efficiency and a native 262K context window.

发布日期
2026-02-24
参数规模
35.0B
上下文长度
262K
支持模态
audio, image, text, video

能力雷达图

21
general
29
coding
82
reasoning
50
science估算
60
agents
80
multimodal

Science 在缺少专门科学评测时使用推理能力代理估算。

排行榜排名

领域#排名分数来源
智能体能力模型榜53
56.0
LS
代码能力榜221
45.0
AA
通用能力榜183
52.0
AA
多模态榜70
69.0
LS
推理能力57
68.0
LS
科学能力167
52.0
AA

基准测试分数 (LLM Stats)

3d

SUNRGBD0.33 / 100自报
Hypersim0.13 / 100自报

Agents

t2-bench81.2%自报
AndroidWorld_SR71.1%自报
BFCL-V467.3%自报
BrowseComp61.0%自报
FullStackBench en58.1%自报
WideSearch57.1%自报
TIR-Bench55.5%自报
FullStackBench zh55.0%自报
OSWorld-Verified54.5%自报
Terminal-Bench 2.040.5%自报
VITA-Bench31.9%自报
DeepPlanning22.8%自报

Biology

GPQA84.2%自报

Chemistry

SuperGPQA63.4%自报

Code

SWE-Bench Verified69.2%自报

Communication

Multi-Challenge60.0%自报

Embodied

EmbSpatialBench0.83 / 100自报

Finance

MMLU-Pro85.3%自报
MMLU-ProX81.0%自报

General

MMLU-Redux93.3%自报
IFEval91.9%自报
C-Eval90.2%自报
MAXIFE86.6%自报
Global PIQA86.6%自报
MMMLU85.2%自报
MMStar81.9%自报
MMMU81.4%自报
Include79.7%自报
MMMU-Pro75.1%自报
LiveCodeBench v674.6%自报
IFBench70.2%自报
LongBench v259.0%自报
SimpleVQA0.58 / 100自报
NOVA-6357.1%自报

Grounding

RefCOCO-avg0.89 / 100自报
ScreenSpot Pro68.6%自报
RefSpatialBench0.64 / 100自报

Healthcare

VideoMMMU80.4%自报
SlakeVQA78.7%自报
PMC-VQA62.0%自报
MedXpertQA61.4%自报

Image To Text

OCRBench91.0%自报

Language

LingoQA79.2%自报
WMT24++76.3%自报

Long Context

MLVU85.6%自报
LVBench71.4%自报
MMLongBench-Doc0.59 / 100自报
AA-LCR58.5%自报

Math

HMMT2589.2%自报
HMMT 202589.0%自报
MathVista-Mini86.2%自报
DynaMath85.0%自报
MathVision83.9%自报
CodeForces0.82 / 3000自报
PolyMATH64.4%自报
Humanity's Last Exam47.4%自报

Multimodal

VLMsAreBlind97.0%自报
V*92.7%自报
AI2D92.6%自报
MMBench-V1.191.5%自报
OmniDocBench 1.589.3%自报
VideoMME w sub.86.6%自报
VideoMME w/o sub.82.5%自报
CC-OCR80.7%自报
CharXiv-R77.5%自报
MVBench74.8%自报
MMVU72.3%自报
BabyVision38.4%自报
ZEROBench-Sub0.34 / 100自报
Nuscene14.6%自报
ZEROBench0.08 / 100自报

Reasoning

CountBench0.98 / 100自报
BrowseComp-zh69.5%自报
Hallusion Bench67.9%自报
ERQA64.8%自报
Seal-041.4%自报
OJBench36.0%自报

Spatial Reasoning

RealWorldQA84.1%自报

Vision

ODinW42.6%自报

AA 评测指数

Intelligence Index
23.4
Tau2
0.9
Gpqa
0.8
Lcr
0.6
Ifbench
0.4
Scicode
0.3
Hle
0.1
Terminalbench Hard
0.1

LLM Stats 分类评分

Image To Text
80
Instruction Following
80
Language
80
Legal
80
Math
80
Physics
80
Structured Output
80
Embodied
80
Finance
80
General
80
Biology
80
Text-to-image
80
Video
80
Long Context
70
Multimodal
70
Reasoning
70
Spatial Reasoning
70
Frontend Development
70
Grounding
70
Healthcare
70
Chemistry
70
Vision
70
Search
60
Code
60
Communication
60
Economics
60
Tool Calling
60
Agents
50
3d
20
Spatial
10

定价

输入价格$0.25 / 1M tokens
输出价格$2 / 1M tokens
混合价格(3:1)$0.688 / 1M tokens

速度

Tokens/秒203.7
首Token延迟1.26s
首回答延迟1.26s

供应商价格排行

供应商价格排行

12 个供应商

最便宜: Novita最贵: Venice AI
供应商输入输出
1Novita最便宜
$0
$0
2OrcaRouter
$0.057
$0.459
3OpenRouter
$0.14
$1
4Deep Infra
$0.14
$1
5Kilo Gateway
$0.1625
$1.3
6NanoGPT
$0.225
$1.8
7SiliconFlow (China)
$0.23
$1.86
8Alibaba主要
$0.25
$2
9Hugging Face
$0.25
$2
10NovitaAI
$0.25
$2
11Mixlayer
$0.25
$1.3
12Venice AI
$0.3125
$1.25

比较该模型在不同 API 供应商之间的定价。

外部链接

LLM StatsArtificial Analysis