Hesam(@Hesamation)

well if you were wondering why the Chinese labs have been silent lately, here you go: the 2.4 TRILLION param Qwen4-Max. it actually beats Opus 5.5 on DeepSWE and Agents Last Exam in Qwen’s evals, and costs just $1.50/$4.50 per 1M input/output tokens, compared to Opus’s $4/$20. it also falls behind substantially in some other benchmarks so take the numbers with a grain of salt, but compared to Qwen3.8-Max, it’s a giant leap. I just hope they drop a 27B soon too.

8.5内容质量
well if you were wondering why the Chinese labs have been silent lately, here you go: the 2.4 TRILLION param Qwen4-Max.

it actually beats Opus 5.5 on DeepSWE and Agents Last Exam in Qwen’s evals, and costs just $1.50/$4.50 per 1M input/output tokens, compared to Opus’s $4/$20.

it also falls behind substantially in some other benchmarks so take the numbers with a grain of salt,

but compared to Qwen3.8-Max, it’s a giant leap.

I just hope they drop a 27B soon too.

TL;DR · AI 摘要

Qwen4-Max 是一个参数量达 2.4 万亿的模型,在部分基准测试中表现优于 Opus 5.5,且成本更低。

核心要点

  • Qwen4-Max 在 DeepSWE 和 Agents Last Exam 基准测试中表现优于 Opus 5.5。
  • Qwen4-Max 的输入/输出每百万 token 成本为 $1.50/$4.50,远低于 Opus 的 $4/$20。
  • Qwen4-Max 相比 Qwen3.8-Max 有显著性能提升,但其他基准测试中表现略逊。

结构提纲

按章节快速跳转。

  1. 介绍 Qwen4-Max 的发布背景及其参数规模。

  2. Qwen4-Max 在 DeepSWE 和 Agents Last Exam 基准测试中表现优于 Opus 5.5。

  3. Qwen4-Max 的输入/输出成本显著低于 Opus。

  4. Qwen4-Max 相比 Qwen3.8-Max 有显著性能提升,但其他基准测试中表现略逊。

  5. 作者希望阿里巴巴能尽快推出 27B 参数规模的模型。

思维导图

用一张图看清主题之间的关系。

查看大纲文本(无障碍 / 无 JS 友好)
  • Qwen4-Max 模型分析
    • 性能表现
      • 优于 Opus 5.5 在 DeepSWE 和 Agents Last Exam
      • 相比 Qwen3.8-Max 有显著提升
    • 成本效益
      • 输入/输出成本 $1.50/$4.50(Opus 为 $4/$20)
    • 未来展望
      • 期待 27B 参数模型的发布

金句 / Highlights

值得收藏与分享的关键句。

#Qwen4-Max#AI模型#成本效益#阿里巴巴#MoE架构
打开原文

ℏεsam on X: "well if you were wondering why the Chinese labs have been silent lately, here you go: the 2.4 TRILLION param Qwen4-Max. it actually beats Opus 5.5 on DeepSWE and Agents Last Exam in Qwen’s evals, and costs just $1.50/$4.50 per 1M input/output tokens, compared to Opus’s $4/$20. it als… / X

ℏεsam

@Hesamation

$

/$

well if you were wondering why the Chinese labs have been silent lately, here you go: the 2.4 TRILLION param Qwen4-Max. it actually beats Opus 5.5 on DeepSWE and Agents Last Exam in Qwen’s evals, and costs just $1.50/$4.50 per 1M input/output tokens, compared to Opus’s $4/$20. it also falls behind substantially in some other benchmarks so take the numbers with a grain of salt, but compared to Qwen3.8-Max, it’s a giant leap. I just hope they drop a 27B soon too.

Qwen

@Alibaba_Qwen

3h

📢 Meet Qwen4-Max, your new AI teammate for work, creativity, and discovery! Qwen4-Max is built on a new MoE architecture with native vision and a 1M-token context window. Highlights: - Outperforms Qwen3.8-Max with up to ~50% fewer tokens, while average cost per task drops over

2:26 AM · Nov 16, 2026

·

12.4K

Views

21

3

125

27