跳转到主要内容
AITier

Qwen3.5 397B A17B (Non-reasoning)

AlibabaQwen开源权重Apache 2.0 · 商用许可

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.

发布日期
2026-02-16
参数规模
397.0B
上下文长度
262K
支持模态
audio, image, text, video

能力雷达图

29
general
41
coding
86
reasoning
57
science估算
60
agents
70
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60
55.0
LS
代码能力榜100
67.0
AA
通用能力榜110
62.0
AA
推理能力69
58.0
LS
科学能力80
64.0
AA

基准测试分数 (LLM Stats)

Agents

GDPval-AA961.00 / 3000自报
t2-bench86.7%自报
WideSearch74.0%自报
BFCL-V472.9%自报
BrowseComp69.0%自报
Terminal-Bench 2.052.5%自报
VITA-Bench49.7%自报
MCP-Mark46.1%自报
Toolathlon38.3%自报
DeepPlanning34.3%自报

Biology

GPQA88.4%自报

Chemistry

SuperGPQA70.4%自报

Code

SWE-Bench Verified76.4%自报
SWE-bench Multilingual69.3%自报

Coding

SecCodeBench68.3%自报

Communication

Multi-Challenge67.6%自报

Finance

MMLU-Pro87.8%自报
MMLU-ProX84.7%自报

General

MMLU-Redux94.9%自报
C-Eval93.0%自报
IFEval92.6%自报
Global PIQA89.8%自报
MMMLU88.5%自报
MAXIFE88.2%自报
Include85.6%自报
LiveCodeBench v683.6%自报
IFBench76.5%自报
LongBench v263.2%自报
NOVA-6359.1%自报

Language

WMT24++78.9%自报

Long Context

AA-LCR68.7%自报

Math

HMMT 202594.8%自报
HMMT2592.7%自报
AIME 202691.3%自报
IMO-AnswerBench80.9%自报
PolyMATH73.3%自报
Humanity's Last Exam28.7%自报

Reasoning

BrowseComp-zh70.3%自报
Seal-046.9%自报

AA 评测指数

Intelligence Index
32.0
Gpqa
0.9
Tau2
0.8
Lcr
0.6
Ifbench
0.5
Scicode
0.4
Terminalbench Hard
0.4
Hle
0.2

LLM Stats 分类评分

Legal
100
Finance
100
Agents
97
General
61
Reasoning
35
Language
90
Biology
90
Instruction Following
80
Math
80
Physics
80
Structured Output
80
Frontend Development
80
Healthcare
80
Chemistry
80
Long Context
70
Multimodal
70
Search
70
Spatial Reasoning
70
Code
70
Communication
70
Economics
70
Tool Calling
60
Vision
50

定价

输入价格$0.6 / 1M tokens
输出价格$3.6 / 1M tokens
混合价格(3:1)$1.35 / 1M tokens

速度

Tokens/秒52.9
首Token延迟1.75s
首回答延迟1.75s

供应商价格排行

供应商价格排行

9 个供应商

最便宜: Novita最贵: OVHcloud AI Endpoints
供应商输入输出
1Novita最便宜
$0
$0
2CrofAI
$0.35
$1.75
3Alibaba (China)
$0.43
$2.58
4DigitalOcean
$0.55
$3.5
5Alibaba主要
$0.6
$3.6
6Scaleway
$0.6
$3.6
7NanoGPT
$0.6
$3.6
8Cortecs
$0.6
$3.6
9OVHcloud AI Endpoints
$0.71
$4.25

比较该模型在不同 API 供应商之间的定价。

外部链接

LLM StatsArtificial Analysis