Qwen3-Next may be Alibaba’s biggest step yet in the AI race

WEB DESK: Alibaba Group has unveiled its latest artificial intelligence model, describing it as one of the company’s most efficient and powerful breakthroughs yet.

The model, called Qwen3-Next-80B-A3B, is built on Alibaba Cloud’s new Qwen3-Next architecture and was made open-source on Friday, giving developers around the world free access to its code and design.

According to Alibaba’s Qwen team, the new model delivers ten times the performance of its predecessor while costing just a tenth as much to build and train.

It is the latest addition to Alibaba’s fast-growing Qwen family of AI models, which have steadily gained attention in the global developer community for their strong performance and open-source approach.

Designed to be faster, cheaper and more accessible

The Qwen3-Next-80B-A3B model has 80 billion parameters, a key measure of an AI model’s complexity and potential capabilities. Despite its size, Alibaba says it runs faster and more efficiently than earlier versions. Compared with the previous Qwen3-32B model launched in April, the new system is significantly quicker in completing tasks while using far fewer resources.

Alibaba Cloud, the company’s cloud computing and AI arm, said the model can also match the performance of its flagship Qwen3-235B-A22B model, which has nearly three times as many parameters.

What sets the new model apart, according to the company, is that it can run on consumer-grade hardware, making it easier and cheaper for smaller developers, startups and even hobbyists to experiment with cutting-edge AI.

The model has been uploaded to both GitHub and Hugging Face, two of the world’s most widely used developer platforms, allowing programmers and AI researchers to test, modify and build upon it freely.

Closing the gap with rivals

The release reflects how Chinese tech firms, including Alibaba, are increasingly using open-source strategies to narrow the gap with US AI giants. By sharing the underlying code, Alibaba hopes to speed up innovation and create a larger ecosystem around its AI tools.

The Qwen team also introduced a specialised reasoning-focused version called Qwen3-Next-80B-A3B-Thinking. According to third-party benchmark tests, this model not only outperformed Alibaba’s own Qwen3-32B-Thinking system but also beat Google’s Gemini-2.5-Flash-Thinking on several reasoning tasks.

Part of a wider AI push

Alibaba has been pushing aggressively into AI development over the past year. In June, the company released open-source Qwen3 models optimised for Apple’s MLX machine learning framework. Earlier reports suggested that Apple had partnered with Alibaba to use its Qwen models for Apple Intelligence services in mainland China, while relying on OpenAI’s GPT models elsewhere.

The Qwen3-Next launch also comes just a week after Alibaba Cloud revealed its largest AI model yet, Qwen-3-Max-Preview, which packs over 1 trillion parameters and debuted in sixth place on UC Berkeley’s LMArena text model rankings.

By releasing high-performing models like Qwen3-Next at a fraction of the cost, Alibaba is signalling its intention to make powerful AI more widely accessible, potentially reshaping how developers and companies build future applications.

Qwen AI’s image generation is quicker but not quite accurate

Now let’s talk a bit about its flaws. Despite all its impressive features and capabilities, the free version of Qwen3 AI currently shows mixed results in image generation. At times, it forgets that it has image generation capabilities altogether, returning errors or saying it cannot do it.

Other times, it generates images within seconds. When it is working properly, its speed does seem slightly faster than ChatGPT, but it still struggles with the same limitations seen in other free AI image tools.

Even when a specific size is requested, it often produces images at 1024×1024 or 1328×1328, ignoring the given dimensions. No matter how detailed the instructions are, it does not always follow them accurately.

Read next: iOS 26 releases today: what’s new and which iPhones won’t get it