---
title: "Alibaba Unveils Qwen4-Max: Flagship 2.4T Parameter AI Model with Built-in Vision and Strict Cybersecurity Constraints"
description: "Alibaba has introduced its flagship Qwen4-Max AI model, featuring 2.4 trillion parameters, built-in vision, and a 1-million token context window, while implementing strict cybersecurity and CBRN-risk constraints."
date: 2026-11-16T19:10:26.471Z
lang: en
url: https://xab.info/en/posts/alibaba-unveils-qwen4-max-flagship-ai-model
tags: [alibaba, qwen4-max, ai-model, cybersecurity, large-language-model]
author: "Alexander Buran"
publisher: "XAB.info"
---

# Alibaba Unveils Qwen4-Max: Flagship 2.4T Parameter AI Model with Built-in Vision and Strict Cybersecurity Constraints

![Qwen 4 logo on a white background with text about the new AI model launch](https://xab.info/media/2026/11/16/alibaba-predstavila-qwen4-max-flagmanskaya-ii-model/alibaba-predstavila-qwen4-max-flagmanskaya-ii-model-1.webp)

## Key Points

- Qwen4-Max contains 2.4 trillion parameters but activates only 71 billion per token—25% less than Qwen3.8-Max.
- The model supports image processing and has a 1 million token context window.
- Alibaba filtered training data for 'offensive' cyber capabilities and introduced datasets to mitigate CBRN risks.
- Task costs in the Work360 benchmark have decreased by 60%; API pricing ranges from $1.5 to $4.5 per 1 million tokens.
- Qwen4-Max will remain closed, but smaller versions with open weights are coming soon.

On November 16, 2026, Alibaba officially introduced Qwen4-Max, a new flagship neural network model representing the next stage in the Qwen family lineage. With a total parameter count of 2.4 trillion, the algorithm matches its predecessor, Qwen3.8-Max; however, the fundamental difference lies in its activation architecture: the model utilizes only 71 billion parameters per token during generation. This is 25% less than its predecessor, allowing the developer to maintain high reasoning capabilities while significantly reducing inference costs. Beyond its text core, Qwen4-Max features built-in image processing support and an expanded context window of 1 million tokens, opening up new possibilities for long-document analysis and multimodal tasks.

### Economic Efficiency and Pricing Structure

A key argument from Alibaba in favor of the new model is its economic efficiency. According to the company, the average cost per task in Work360—an internal benchmark for evaluating model performance in applied scenarios—has decreased by 60% compared to the previous version. For developers and corporate clients, Alibaba has released a transparent API pricing structure: cached input is priced at $0.075 per 1 million tokens, input tokens at $1.5 per 1 million, and output tokens at $4.5 per 1 million. This combination of reduced inference costs and expanded context makes the model highly competitive in tasks where both data volume and predictable spending are critical.

### Cybersecurity and CBRN Constraints as a Systemic Priority

A separate part of the announcement focused on security measures. Alibaba stated that it intentionally filtered its training data to remove information that could be used for "offensive" cybersecurity actions while retaining useful defensive functions, specifically vulnerability scanning capabilities. In addition, the company introduced specialized datasets for fine-tuning aimed at mitigating risks related to chemical, biological, radiological, and nuclear (CBRN) weapons. Observers note that this stance reflects Alibaba's attempt to balance its commitment to open source with growing global concerns about AI-powered cyberattacks, alongside a heightened focus on safety across the entire Chinese AI industry.

### Scientific Results: From Refrigerants to Alzheimer's

Alibaba also highlighted the applied scientific achievements of Qwen4-Max. Specifically, the company reported on a continuous 48-hour autonomous research project during which a new-generation refrigerant molecule for semiconductor production was discovered and experimentally confirmed. A separate autonomous study focused on Alzheimer's disease is also mentioned. The developer uses these cases to demonstrate the model's ability to conduct long, multi-step research processes without constant external intervention.

### Openness Status and Roadmap for Weights

Although Qwen4-Max itself remains a closed model, Alibaba confirmed that smaller Qwen4 versions with open weights will become available in the near future. Thus, the company is maintaining a two-tier strategy: a flagship closed product for corporate and high-load scenarios, and open versions for the developer community, which supports the ecosystem and lowers the barrier to entry for third-party teams.

## Fact-Check & Sources

- [Alibaba представила Qwen4-Max — флагманскую ИИ-модель с  2,4 трлн параметров и запретом на кибератаки](https://3dnews.ru/1151354) - Source: parameters, 71B activation, 1M context, pricing, Work360 -60%, CBRN datasets, research cases, weight status.
- [Alibaba представила Qwen4-Max — флагманскую ИИ-модель с 2,4 трлн параметров и запретом на кибератаки](https://3dnews.ru/1151354/alibaba-predstavila-qwen4max-flagmanskuyu-iimodel-s-24-trln-parametrov-i-zapretom-na-kiberataki) - Первоисточник: параметры, активация 71 млрд, контекст 1 млн, тарифы, Work360 -60%, CBRN-наборы данных, научные кейсы, статус весов.
- [Alibaba представила Qwen4-Max с 2,4 трлн параметров и ограничениями в кибербезопасности](https://runet.news/news/89730) - Независимое пересказывание: параметры, зрение, фильтрация данных, CBRN, научные исследования, открытые веса.
- [Alibaba выпустила Qwen4-Max: что умеет новая флагманская нейросеть из Китая](https://newizv.ru/news/2026-11-16/alibaba-vypustila-qwen4-max-chto-umeet-novaya-flagmanskaya-neyroset-iz-kitaya-447645) - Подтверждает дату 16.11.2026, контекст 1 млн, зрение; указывает снижение стоимости рабочих задач на 62% (против 60% у 3dnews).

## FAQ

### How many parameters does Qwen4-Max have, and how many are active?

The model has a total of 2.4 trillion parameters, but only 71 billion are activated per token—25% fewer than the previous Qwen3.8-Max.

### What is the API pricing for Qwen4-Max?

According to Alibaba, cached input costs $0.075 per 1 million tokens, input tokens cost $1.5 per 1 million, and output tokens cost $4.5 per 1 million.

### Will Qwen4-Max be open source?

Qwen4-Max itself will remain a closed model, but Alibaba has confirmed that smaller Qwen4 versions with open weights will be released soon.

### What security measures has Alibaba introduced?

The company filtered training data to exclude 'offensive' cyber capabilities, maintained defensive functions, and introduced datasets to mitigate CBRN weapon risks.