跳转到主要内容

Qwen3 Max Thinking

AlibabaQwenProprietary

描述

The latest flagship reasoning model in the Qwen3 family. Further enhanced by multiple innovations like adaptive tool-use and advanced test-time scaling techniques

发布日期
2026-01-26
参数规模
1.0T
上下文长度
262K
支持模态
text

能力雷达图

60
general
50
coding
70
reasoning
77
science估算
40
agents
0
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

领域#排名分数来源
智能体能力模型榜109
41.0
LS

基准测试分数 (LLM Stats)

(LLM Stats (zeroeval))

Agents

VITA-Bench40.9%自报
MCP-Mark33.5%自报
DeepPlanning28.7%自报
Toolathlon18.8%自报

Chat

IFEvalGoogle Research (2023)93.4%自报
Multi-Challenge63.3%自报

Code

SecCodeBench57.5%自报

General

C-Eval93.7%自报
MAXIFE84.0%自报
NOVA-6354.2%自报

Instruction Following

IFBench70.9%自报

Knowledge

HLE-Verified37.6%自报

Language

MMLU-Redux92.8%自报
MMLU-Pro85.7%自报
MMMLU84.4%自报
WMT24++77.6%自报

Long Context

LongBench v260.6%自报

Math

AIME 202693.3%自报
IMO-AnswerBench83.9%自报

Reasoning

GPQANYU + Cohere + Anthropic (2023)87.4%自报
Global PIQA86.0%自报
LiveCodeBench v685.9%自报
SWE-Bench Verified75.3%自报
AA-LCR68.7%自报
SuperGPQA67.3%自报
SWE-bench Multilingual66.7%自报
BrowseComp-zh60.9%自报
BrowseCompOpenAI (2025)53.9%自报
Humanity's Last Exam (with tools, text-only)49.8%自报
Seal-046.9%自报
Humanity's Last Exam (no tools, text-only)30.2%自报
Terminal-Bench 2.0Stanford × Laude Institute (2026)22.5%自报

Search

WideSearch57.9%自报

Tool Calling

BFCL-V467.7%自报

AA 评测指数

(Artificial Analysis)

暂无 AA 评测数据

LLM Stats 分类评分

(LLM Stats (zeroeval))
Language
90
Biology
90
Chat
80
Instruction Following
80
Legal
80
Physics
80
Structured Output
80
Finance
80
Frontend Development
80
Healthcare
80
Chemistry
80
Math
70
Economics
70
Long Context
60
Reasoning
60
General
60
Communication
60
Search
50
Code
50
Knowledge
40
Agents
40
Tool Calling
40

定价

输入价格$1.2 / 1M tokens
输出价格$6 / 1M tokens
混合价格(3:1)$2.4 / 1M tokens

速度

暂无速度数据

供应商价格排行

供应商价格排行

2 个供应商

最便宜: Alibaba最贵: DeepInfra
供应商输入输出
1Alibaba主要
$1.2
$6
2DeepInfra
$1.2
$6

比较该模型在不同 API 供应商之间的定价。

外部链接