Qwen3.5 397B A17B (Non-reasoning)

AlibabaQwen開源權重Apache 2.0 · 商用許可

描述

Qwen3.5-397B-A17B is Qwen's flagship Mixture-of-Experts model with 397 billion total parameters and 17 billion activated parameters. It delivers state-of-the-art performance across knowledge, reasoning, coding, mathematics, multilingual understanding, instruction following, long context, and agent tasks.

發布日期

2026-02-16

參數規模

397.0B

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	17	62.0	LS
程式碼能力榜	216	64.0	AA
通用能力榜	211	52.0	AA
科學能力	195	60.0	AA

基準測試分數 (LLM Stats)

(LLM Stats (zeroeval))

Agents

t2-bench

86.7%自報

VITA-Bench

49.7%自報

MCP-Mark

46.1%自報

Toolathlon

38.3%自報

DeepPlanning

34.3%自報

Chat

IFEvalGoogle Research (2023)

92.6%自報

Multi-Challenge

67.6%自報

Code

SecCodeBench

68.3%自報

General

C-Eval

93.0%自報

MAXIFE

88.2%自報

Include

85.6%自報

NOVA-63

59.1%自報

Instruction Following

IFBench

76.5%自報

Language

MMLU-Redux

94.9%自報

MMMLU

88.5%自報

MMLU-Pro

87.8%自報

MMLU-ProX

84.7%自報

WMT24++

78.9%自報

Long Context

LongBench v2

63.2%自報

Math

HMMT 2025

94.8%自報

HMMT25

92.7%自報

AIME 2026

91.3%自報

IMO-AnswerBench

80.9%自報

PolyMATH

73.3%自報

Reasoning

Global PIQA

89.8%自報

GPQANYU + Cohere + Anthropic (2023)

88.4%自報

LiveCodeBench v6

83.6%自報

SWE-Bench Verified

76.4%自報

SuperGPQA

70.4%自報

BrowseComp-zh

70.3%自報

SWE-bench Multilingual

69.3%自報

BrowseCompOpenAI (2025)

69.0%自報

AA-LCR

68.7%自報

Terminal-Bench 2.0Stanford × Laude Institute (2026)

52.5%自報

Seal-0

46.9%自報

Humanity's Last Exam

28.7%自報

Search

WideSearch

74.0%自報

Tool Calling

BFCL-V4

72.9%自報

AA 評測指數

(Artificial Analysis)

Gpqa(NYU + Cohere + Anthropic (2023))

86.1

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

83.9

Lcr(Artificial Analysis)

64.3

Ifbench(Google Research (2023))

51.6

Terminalbench Hard(Stanford × Laude Institute (2026))

35.6

Intelligence Index(Artificial Analysis)

21.4

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

19.8

LLM Stats 分類評分

(LLM Stats (zeroeval))

Language

Biology

Chat

Instruction Following

Legal

Math

Physics

Structured Output

Finance

Frontend Development

Healthcare

Chemistry

Long Context

Multimodal

Reasoning

Spatial Reasoning

General

Code

Communication

Economics

Agents

Tool Calling

Vision

定價

輸入價格$0.6 / 1M tokens

輸出價格$3.6 / 1M tokens

混合價格(3:1)$1.35 / 1M tokens

速度

Tokens/秒86.1

首Token延遲1.70s

首回答延遲1.70s

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

$0.6

$3.6

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