Alibaba’s Qwen3 AI model coming this month, sources say, in bid to cement industry lead

Alibaba Group Holding is preparing to launch the next generation of its flagship Qwen series of large language models (LLMs) in the coming weeks, aiming to solidify its position as a leader in open-source artificial intelligence (AI).

Alibaba Cloud, the cloud computing unit of the Hangzhou-based tech giant, plans to launch its Qwen3 LLM this month, according to people familiar with the matter. Alibaba owns the South China Morning Post.

The upgraded model is expected to come in multiple variants, including a standard version and a Qwen3-MoE, which uses a mixture-of-experts architecture, one person with knowledge of the plans said. The MoE design allows for cost-efficient training and deployment, a strategy that helped DeepSeek develop its V3 model at a fraction of the cost of leading US models.

With Qwen3, Alibaba seeks to cement its role as a leading provider of open-source models. Last Wednesday, the company released Qwen2.5-Omni-7B, a multimodal model capable of processing text, images, audio and video. It quickly became the most popular model on Hugging Face, the world’s largest open-source AI community, followed by DeepSeek’s updated V3 model.

Alibaba has not given an official timeline for the release of the new model. The company did not respond to a request for comment on Wednesday.

Alibaba’s push also comes amid intensifying competition among Big Tech firms as they race to develop and upgrade their own foundation models. While only a handful of companies can afford to develop in-house general-purpose LLMs, many smaller companies have shifted to focus on building specialised applications around publicly available models.