跳轉到主要內容

Qwen3 Next 80B A3B (Reasoning)

AlibabaQwen開源權重Apache 2.0 · 商用許可

描述

Qwen3-Next-80B-A3B-Thinking is the thinking variant of the Qwen3-Next series, featuring the same groundbreaking architecture as the instruct model. Leveraging GSPO, it addresses stability and efficiency challenges of hybrid attention + high-sparsity MoE in RL training. It uses Hybrid Attention combining Gated DeltaNet and Gated Attention for efficient ultra-long context modeling, High-Sparsity MoE with 512 experts (10 activated + 1 shared), and Multi-Token Prediction. With 80B total parameters and only 3B activated, it demonstrates outstanding performance on complex reasoning tasks — outperforming Qwen3-30B-A3B-Thinking-2507, Qwen3-32B-Thinking, and even the proprietary Gemini-2.5-Flash-Thinking across multiple benchmarks. Architecture: 48 layers, 15T training tokens, hybrid layout of 12*(3*(Gated DeltaNet->MoE)->(Gated Attention->MoE)). Supports only thinking mode with automatic <think> tag inclusion, may generate longer thinking content.

發布日期
2025-09-11
參數規模
80.0B
上下文長度
131K
支援模態
text

能力雷達圖

70
general
0
coding
80
reasoning
68
science估算
60
agents
0
multimodal

缺少專門科學評測時,Science 由 LLM Stats 科學得分或推理能力估算。

排行榜排名

暫無排名資料

基準測試分數 (LLM Stats)

(LLM Stats (zeroeval))

Chat

IFEvalGoogle Research (2023)88.9%自報
Multi-IF77.8%自報
TAU-bench Retail69.6%自報
Tau2 Airline60.5%自報

Code

CFEval2071.00 / 10000自報

Communication

Tau2 Retail67.8%自報
Tau2 Telecom43.9%自報

General

Include78.9%自報
BFCL-v372.0%自報
Arena-Hard v262.3%自報

Language

MMLU-Redux92.5%自報
MMLU-Pro82.7%自報
MMLU-ProX78.7%自報

Math

AIME 202587.8%自報
LiveBench 2024112576.6%自報
HMMT2573.9%自報
PolyMATH56.3%自報

Reasoning

GPQANYU + Cohere + Anthropic (2023)77.2%自報
LiveCodeBench v668.7%自報
SuperGPQA60.8%自報
TAU-bench Airline49.0%自報
OJBench29.7%自報

Writing

WritingBench84.6%自報

AA 評測指數

(Artificial Analysis)

暫無 AA 評測資料

LLM Stats 分類評分

(LLM Stats (zeroeval))
Instruction Following
80
Language
80
Legal
80
Math
80
Structured Output
80
Finance
80
Biology
80
Chat
70
Physics
70
Reasoning
70
General
70
Healthcare
70
Agents
70
Chemistry
70
Creativity
70
Writing
70
Multimodal
60
Spatial Reasoning
60
Communication
60
Economics
60
Tool Calling
60
Vision
60

定價

輸入價格$0.5 / 1M tokens
輸出價格$6 / 1M tokens
混合價格(3:1)$1.875 / 1M tokens

速度

暫無速度資料

供應商價格排行

供應商價格排行

3 個供應商

最便宜: Alibaba (China)最貴: Alibaba
供應商輸入輸出
1Alibaba (China)最便宜
$0.144
$1.434
2Qiniu
$0.14493
$1.44928
3Alibaba主要
$0.5
$6

比較該模型在不同 API 供應商之間的定價。

外部連結