---
title: "Alibaba releases Qwen4-Max, a 2.4-trillion-parameter proprietary model"
url: https://www.parallelquant.com/posts/alibaba-releases-qwen4-max-a-2-4-trillion-parameter-proprietary-model-7a59ae
source_name: "MarkTechPost"
source_url: https://www.marktechpost.com/2026/11/16/alibaba-qwen4-max-release/
published: 2026-11-16T09:44:17.000Z
topics: ["llms", "products"]
publisher: "Parallel Quant"
---

# Alibaba releases Qwen4-Max, a 2.4-trillion-parameter proprietary model

*2026-11-16 · Source: [MarkTechPost](https://www.marktechpost.com/2026/11/16/alibaba-qwen4-max-release/)*

Alibaba released Qwen4-Max, a proprietary mixture-of-experts (MoE) model with 2.4 trillion total parameters and 71 billion active per token. It has native vision, a 1-million-token context window, and costs $1.5 per million input tokens. Alibaba says it cuts cost per task on its Work360 benchmark by 62%.

**Why it matters:** Keeping the flagship closed while open-sourcing smaller Qwen models follows the same split strategy US labs use, and suggests Chinese labs now see their top tier as worth protecting commercially. Coming days after Meituan's 1.6-trillion-parameter open agent model, it shows Chinese frontier scale is advancing despite the ongoing fight over Nvidia chip exports. Low input pricing at this size also keeps pressure on Western labs' API margins.

**Topics:** llms, products

---
Read the original: https://www.marktechpost.com/2026/11/16/alibaba-qwen4-max-release/
Canonical: https://www.parallelquant.com/posts/alibaba-releases-qwen4-max-a-2-4-trillion-parameter-proprietary-model-7a59ae
Published by Parallel Quant — https://www.parallelquant.com
