Alibaba Opens Qwen-Max to Open Source, Price of Closed AI Model API Can Be Disrupted

Alibaba logo. Alibaba will open its flagship AI model, Qwen-Max, as open source for the first time. Photo: Yicai Global.

JAKARTA - Alibaba is changing its big strategy in the artificial intelligence business. The Chinese internet giant will open Qwen-Max, its flagship AI model, as an open source model for the first time.

Yicai Global, quoted Thursday, August 6, reported that the decision came after Chinese AI companies such as DeepSeek showed that open source and low-cost strategies could form their own markets.

Alibaba released the new generation of the basic model of Qwen3.8-Max on August 3. This model has 2.4 trillion parameters.

In AI, parameters are components that help the model recognize patterns, understand commands, make answers, and perform tasks.

Alibaba claims that Qwen3.8-Max is one of the best large-scale language models in the world in terms of overall performance. The claim includes programming, scientific research, and long-term tasks.

The most important change is in its business strategy. Alibaba plans to open the weight of the Qwen3.8-Max model as open source next week.

The weight of the model is the main content that makes AI work and give answers. If the weight is opened, developers can download, study, and run the model more freely.

Since launching the Qwen series in 2023, Alibaba has released several open source models. However, the Max series has previously always been positioned as a closed flagship model.

The closed model can only be accessed via an application programming interface or API. An API is a path that other applications use to connect and use the capabilities of the AI model. Users can't download the model weights.

This move shows that Alibaba is no longer solely based on the potential for profit from closed APIs. The Hangzhou-based company chose to expand its influence through open source channels, especially for the AI agent ecosystem.

AI agents are systems that can carry out certain tasks more independently based on user instructions.

A number of analysts believe that the open source strategy could hinder commercialization. However, this strategy also lowers the barriers to adoption, expands its reach abroad, and encourages cross-cloud use and various types of hardware.

Robust open source alternatives could also make closed model providers more difficult to maintain high API prices.

Yicai Global noted that the cost of Qwen3.8-Max was around 6 US dollars per one million output tokens. Meanwhile, the cost of Anthropic's Claude Fable 5 model reached 50 US dollars.

Tokens are pieces of text that the AI model processes. Token costs are important because the workflow of an AI agent usually requires repeated tool calls, long code generation, feedback many times, and a continuous trial-and-error process.

In such a need, the cost difference can immediately determine whether a model is worth using commercially or not.

Even so, the computing power ecosystem is still a weak point for AI startups such as DeepSeek. Computing power is needed to run AI models on a large scale, especially when the number of users surges.

The overseas open source community OpenCode said DeepSeek V4 Flash is experiencing capacity problems due to an unprecedented traffic surge.

"Users may experience an error, and urgent repairs are being carried out," OpenCode said.