# Alibaba Launches New Generation Base Large Model Qwen4-Max

- Category: Flash (快讯)
- Published: 2026-11-16T11:58:00.000Z
- Source date: 2026-11-16
- HTML: https://www.88z.ai/en/news/flash-20261116-8e5b7ab7a05e51e8
- Markdown: https://www.88z.ai/en/news/flash-20261116-8e5b7ab7a05e51e8.md
- Other language: https://www.88z.ai/news/flash-20261116-8e5b7ab7a05e51e8.md

Alibaba has launched Qwen4-Max, its new generation base large model, now available on the Qianwen AI platform with API services and Token Plans. Featuring a QSA sparse attention MoE architecture with 2.4 trillion parameters, it supports million-level context windows and native vision. The model shows enhanced coding and agent capabilities, reducing token usage by nearly 50% and increasing scores by almost 28% in Terminal-Bench 4.0 benchmarks, while cutting single-task costs by 58%.

## At a glance

- Qwen4-Max uses QSA sparse attention MoE architecture, supporting million-level context and native vision
- Terminal-Bench 4.0 tests show nearly 50% less token usage and 58% lower single-task costs

## The update

Alibaba has officially released Qwen4-Max, its latest generation base large model, which is now live on the Qianwen AI platform with immediate access to API services and Token Plan subscriptions. Built on an advanced QSA sparse attention Mixture of Experts (MoE) architecture, Qwen4-Max features a total of 2.4 trillion parameters and supports million-level long context windows along with native visual understanding capabilities. In terms of performance, Qwen4-Max significantly enhances coding development and agent collaboration abilities. Benchmark data indicates that under equivalent configurations in Terminal-Bench 4.0, the model reduces token consumption by nearly 50% and increases overall scores by almost 28%, while substantially lowering single-task processing costs by 58%, thereby improving both efficiency and cost-effectiveness.

## Sources

- [千问大模型](https://mp.weixin.qq.com/s/pJUt0xwD6H-WMyzd8TjZaQ) (2026-11-16)

This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
