Qwen3.5 35B A3B (Non-reasoning)

AlibabaQwen開源權重Apache 2.0 · 商用許可

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.

發布日期

2026-02-24

參數規模

35.0B

上下文長度

262K

支援模態

audio, image, text, video

能力雷達圖

general

coding

reasoning

science

agents

multimodal

排行榜排名

領域	#排名	分數	來源
智慧體能力模型榜	96	40.0	LS
程式碼能力榜	256	46.0	AA
通用能力榜	226	52.0	AA
多模態榜	45	47.0	LS
科學能力	227	52.0	AA

基準測試分數 (LLM Stats)

(LLM Stats (zeroeval))

3d

SUNRGBD

0.33 / 100自報

Hypersim

0.13 / 100自報

Agents

t2-bench

81.2%自報

AndroidWorld_SR

71.1%自報

BFCL-V4

67.3%自報

BrowseCompOpenAI (2025)

61.0%自報

FullStackBench en

58.1%自報

WideSearch

57.1%自報

TIR-Bench

55.5%自報

FullStackBench zh

55.0%自報

OSWorld-Verified

54.5%自報

Terminal-Bench 2.0Stanford × Laude Institute (2026)

40.5%自報

VITA-Bench

31.9%自報

DeepPlanning

22.8%自報

Biology

GPQANYU + Cohere + Anthropic (2023)

84.2%自報

Chemistry

SuperGPQA

63.4%自報

Code

SWE-Bench Verified

69.2%自報

Communication

Multi-Challenge

60.0%自報

Embodied

EmbSpatialBench

0.83 / 100自報

Finance

MMLU-Pro

85.3%自報

MMLU-ProX

81.0%自報

General

MMLU-Redux

93.3%自報

IFEvalGoogle Research (2023)

91.9%自報

C-Eval

90.2%自報

MAXIFE

86.6%自報

Global PIQA

86.6%自報

MMMLU

85.2%自報

MMStar

81.9%自報

MMMU

81.4%自報

Include

79.7%自報

MMMU-Pro

75.1%自報

LiveCodeBench v6

74.6%自報

IFBench

70.2%自報

LongBench v2

59.0%自報

SimpleVQA

0.58 / 100自報

NOVA-63

57.1%自報

Grounding

RefCOCO-avg

0.89 / 100自報

ScreenSpot Pro

68.6%自報

RefSpatialBench

0.64 / 100自報

Healthcare

VideoMMMU

80.4%自報

SlakeVQA

78.7%自報

PMC-VQA

62.0%自報

MedXpertQA

61.4%自報

Image To Text

OCRBench

91.0%自報

Language

LingoQA

79.2%自報

WMT24++

76.3%自報

Long Context

MLVU

85.6%自報

LVBench

71.4%自報

MMLongBench-Doc

0.59 / 100自報

AA-LCR

58.5%自報

Math

HMMT25

89.2%自報

HMMT 2025

89.0%自報

MathVista-Mini

86.2%自報

DynaMath

85.0%自報

MathVision

83.9%自報

CodeForces

0.82 / 3000自報

PolyMATH

64.4%自報

Humanity's Last Exam

47.4%自報

Multimodal

VLMsAreBlind

97.0%自報

92.7%自報

AI2D

92.6%自報

MMBench-V1.1

91.5%自報

OmniDocBench 1.5

89.3%自報

VideoMME w sub.

86.6%自報

VideoMME w/o sub.

82.5%自報

CC-OCR

80.7%自報

CharXiv-R

77.5%自報

MVBench

74.8%自報

MMVU

72.3%自報

BabyVision

38.4%自報

ZEROBench-Sub

0.34 / 100自報

Nuscene

14.6%自報

ZEROBench

0.08 / 100自報

Reasoning

CountBench

0.98 / 100自報

BrowseComp-zh

69.5%自報

Hallusion Bench

67.9%自報

ERQA

64.8%自報

Seal-0

41.4%自報

OJBench

36.0%自報

Spatial Reasoning

RealWorldQA

84.1%自報

Vision

ODinW

42.6%自報

AA 評測指數

(Artificial Analysis)

Coding Index(Artificial Analysis)

37.0

Intelligence Index(Artificial Analysis)

24.3

Tau2(Sierra + U Toronto + Vector Institute (2025))

0.9

Gpqa(NYU + Cohere + Anthropic (2023))

0.8

Lcr(Artificial Analysis)

0.6

Ifbench(Google Research (2023))

0.4

Terminalbench V2 1

0.4

Scicode(UIUC + Argonne National Lab (2024))

0.3

Hle(Center for AI Safety + Scale AI (2025))

0.1

Terminalbench Hard(Stanford × Laude Institute (2026))

0.1

Tau Banking

0.0

LLM Stats 分類評分

(LLM Stats (zeroeval))

Legal

Math

Physics

Structured Output

Image To Text

Instruction Following

Language

Embodied

Finance

Biology

Text-to-image

Video

Long Context

Multimodal

Reasoning

Spatial Reasoning

Frontend Development

General

Grounding

Healthcare

Chemistry

Vision

Code

Communication

Economics

Tool Calling

Agents

Spatial

定價

輸入價格$0.25 / 1M tokens

輸出價格$2 / 1M tokens

混合價格(3:1)$0.688 / 1M tokens

速度

Tokens/秒161.3

首Token延遲1.21s

首回答延遲1.21s

供應商價格排行

1 個供應商

供應商輸入輸出

1Alibaba主要

$0.25

比較該模型在不同 API 供應商之間的定價。

外部連結

LLM Stats Artificial Analysis