Qwen3.8 Max (0902)
AlibabaQwen开源权重Qwen3.8-Max License · 商用许可
描述
Qwen3.8 Max, released as the Qwen3.8-2.4T-A95B open-weight checkpoint, is Qwen's flagship mixture-of-experts model for coding, research, professional workflows, and long-horizon agents. It has 2.4 trillion total parameters with 95 billion active parameters, a native 262,144-token context window extensible to about 1 million tokens, and always-on thinking. QwenCloud serves the same model in the Qwen3.8 Max API family with text and image input.
发布日期
2026-09-02
参数规模
2.4T
上下文长度
1.0M
支持模态
image, pdf, text, video
能力雷达图
34
general
70
coding
50
reasoning
77
science估算
60
agents
90
multimodal
缺少专门科学评测时,Science 由 LLM Stats 科学得分或推理能力估算。
排行榜排名
基准测试分数 (LLM Stats)
(LLM Stats (zeroeval))Agents
QwenSVG
1713.00 / 2000自报
AndroidWorld
85.3%自报
MobileWorld
77.8%自报
AndroidBench
75.1%自报
CoWorkBench
74.8%自报
Toolathlon
72.5%自报
Workspace Bench
67.7%自报
Job Bench
53.4%自报
OneMillion Bench
52.5%自报
Agents' Last Exam
52.4%自报
MLS-Bench Lite
41.0%自报
AutomationBench
27.3%自报
Code
QwenReactBench
1724.00 / 2000自报
PaperBench
93.0%自报
QwenSWEBench
80.7%自报
FrontierSWE
73.5%自报
SkillsBench
70.2%自报
Vision2Web
69.0%自报
QwenQoderBench
58.4%自报
DeepSWE 1.1
56.6%自报
NL2Repo
55.9%自报
Healthcare
HealthBench
60.2%自报
Instruction Following
IFBench
82.8%自报
Knowledge
PLawBench
73.2%自报
PRBench-Finance
58.3%自报
PRBench-Legal
57.6%自报
Long Context
MRCR v2 (8-needle)
92.9%自报
LongBench v2
66.3%自报
Multimodal
OSWorld-Verified
86.1%自报
Reasoning
GPQANYU + Cohere + Anthropic (2023)
92.6%自报
Terminal-Bench 2.1
86.6%自报
SWE-Bench ProPrinceton NLP (2024)
67.7%自报
Humanity's Last Exam (with tools, text-only)
56.2%自报
Humanity's Last Exam
43.6%自报
Search
WideSearch
81.9%自报
Vision
VideoMME w sub.
90.4%自报
RealWorldQA
88.0%自报
ScreenSpot Pro
84.5%自报
MMMU-Pro
82.3%自报
LVBench
81.8%自报
ERQA
77.8%自报
PerceptionBench
63.5%自报
AA 评测指数
(Artificial Analysis)Strategy And Ops47.3
Finance And Accounting45.0
Economics42.5
Legal41.0
Healthcare And Medical40.8
Engineering(Artificial Analysis)39.3
Intelligence Index(Artificial Analysis)33.7
LLM Stats 分类评分
(LLM Stats (zeroeval))Physics90
Biology90
Chemistry90
Video90
Instruction Following80
Multimodal80
Search80
Spatial Reasoning80
Grounding80
Vision80
Long Context70
Reasoning70
Structured Output70
General70
Agents70
Code70
Productivity60
Healthcare60
Tool Calling60
Math50
定价
输入价格$2 / 1M tokens
输出价格$6 / 1M tokens
混合价格(3:1)$2.47525 / 1M tokens
缓存读取价格$0.25 / 1M tokens
缓存写入价格$2.5 / 1M tokens
速度
Tokens/秒41.5
首Token延迟2.85s
首回答延迟51.10s
供应商价格排行
供应商价格排行
7 个供应商
最便宜: DeepInfra最贵: Together
供应商输入输出
1DeepInfra最便宜
$1.65
$4.951
2Alibaba主要
$2
$6
3Novita
$2
$6
4Fireworks
$2
$6
5EmpirioLabs AI
$2
$6
6Kenari
$2.4
$6.86
7Together
$2.5
$6.25
比较该模型在不同 API 供应商之间的定价。