Qwen3.5 122B A10B (Non-reasoning)

AlibabaQwen開源權重Apache 2.0 · 商用許可

描述

Qwen3.5-122B-A10B is a multimodal Mixture-of-Experts model with 122 billion total parameters and 10 billion activated parameters. It combines strong reasoning, coding, long-context, 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

122.0B

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	47	49.0	LS
程式碼能力榜	258	55.0	AA
通用能力榜	240	48.0	AA
多模態榜	43	59.0	LS
科學能力	221	55.0	AA

基準測試分數 (LLM Stats)

(LLM Stats (zeroeval))

Agents

t2-bench

79.5%自報

VITA-Bench

33.6%自報

DeepPlanning

24.1%自報

Chat

IFEvalGoogle Research (2023)

93.4%自報

Multi-Challenge

61.5%自報

Code

FullStackBench en

62.6%自報

FullStackBench zh

58.7%自報

General

C-Eval

91.9%自報

MAXIFE

87.9%自報

Include

82.8%自報

AndroidWorld_SR

66.4%自報

NOVA-63

58.6%自報

Healthcare

MedXpertQA

67.3%自報

PMC-VQA

63.3%自報

Instruction Following

IFBench

76.1%自報

Language

MMLU-Redux

94.0%自報

MMMLU

86.7%自報

MMLU-Pro

86.7%自報

MMLU-ProX

82.2%自報

WMT24++

78.3%自報

Long Context

LongBench v2

60.2%自報

Math

HMMT 2025

91.4%自報

HMMT25

90.3%自報

MathVista-Mini

87.4%自報

MathVision

86.2%自報

DynaMath

85.9%自報

CodeForces

0.85 / 3000自報

PolyMATH

68.9%自報

Multimodal

VideoMME w/o sub.

83.9%自報

MMMU

83.9%自報

VideoMMMU

82.0%自報

OSWorld-Verified

58.0%自報

TIR-Bench

53.2%自報

Reasoning

Global PIQA

88.4%自報

GPQANYU + Cohere + Anthropic (2023)

86.6%自報

LiveCodeBench v6

78.9%自報

CharXiv-R

77.2%自報

SWE-Bench Verified

72.0%自報

BrowseComp-zh

69.9%自報

SuperGPQA

67.1%自報

AA-LCR

66.9%自報

BrowseCompOpenAI (2025)

63.8%自報

Terminal-Bench 2.0Stanford × Laude Institute (2026)

49.4%自報

Humanity's Last Exam

47.5%自報

Seal-0

44.1%自報

OJBench

39.5%自報

Search

WideSearch

60.5%自報

Tool Calling

BFCL-V4

72.2%自報

Video

MLVU

87.3%自報

Vision

CountBench

0.97 / 100自報

VLMsAreBlind

96.7%自報

AI2D

93.3%自報

93.2%自報

MMBench-V1.1

92.8%自報

OCRBench

92.1%自報

RefCOCO-avg

0.91 / 100自報

OmniDocBench 1.5

89.8%自報

VideoMME w sub.

87.3%自報

RealWorldQA

85.1%自報

EmbSpatialBench

0.84 / 100自報

MMStar

82.9%自報

CC-OCR

81.8%自報

SlakeVQA

81.6%自報

LingoQA

80.8%自報

MMMU-Pro

76.9%自報

MVBench

76.6%自報

MMVU

74.7%自報

LVBench

74.4%自報

ScreenSpot Pro

70.4%自報

RefSpatialBench

0.69 / 100自報

Hallusion Bench

67.6%自報

ERQA

62.0%自報

SimpleVQA

0.62 / 100自報

MMLongBench-Doc

0.59 / 100自報

ODinW

44.5%自報

BabyVision

40.2%自報

SUNRGBD

0.36 / 100自報

ZEROBench-Sub

0.36 / 100自報

Nuscene

15.4%自報

Hypersim

0.13 / 100自報

ZEROBench

0.09 / 100自報

AA 評測指數

(Artificial Analysis)

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

84.5

Gpqa(NYU + Cohere + Anthropic (2023))

82.7

Lcr(Artificial Analysis)

61.3

Ifbench(Google Research (2023))

50.8

Terminalbench V2 1

47.2

Coding Index(Artificial Analysis)

43.3

Terminalbench Hard(Stanford × Laude Institute (2026))

29.5

Intelligence Index(Artificial Analysis)

17.7

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

15.9

Tau Banking

10.3

LLM Stats 分類評分

(LLM Stats (zeroeval))

Biology

Chat

Image To Text

Instruction Following

Language

Legal

Math

Physics

Structured Output

Embodied

Finance

Grounding

Healthcare

Chemistry

Text-to-image

Video

Long Context

Multimodal

Reasoning

Spatial Reasoning

Frontend Development

General

Economics

Vision

Agents

Code

Communication

Tool Calling

Spatial

定價

輸入價格$0.4 / 1M tokens

輸出價格$3.2 / 1M tokens

混合價格(3:1)$1.1 / 1M tokens

速度

Tokens/秒148.6

首Token延遲1.00s

首回答延遲1.00s

供應商價格排行

2 個供應商

最便宜: DeepInfra最貴: Alibaba

供應商輸入輸出

1DeepInfra最便宜

2Alibaba主要

$0.4

$3.2

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