Alibaba has unveiled a new series of large language models (LLM) called Qwen3, which includes models with parameter counts ranging from 0.6 to 235 billion. Developers claim that the new models demonstrate competitive results in key artificial intelligence tasks, surpassing developments from companies like OpenAI and Google.
This is reported by Finway
Advantages of Qwen3 Models
The Qwen3 models, which include a family of hybrid solutions, implement the MoE (Mixture of Experts) architecture, allowing for efficient task distribution among specialized sub-models. This optimizes resource usage, as the models can handle simple queries and perform complex computations on demand. The total volume of training data used for the models reached approximately 36 trillion tokens, encompassing a variety of training materials, code, Q&A, and synthetic data.
Availability and Market
Currently, the Qwen3 models are already hosted on Hugging Face and GitHub platforms, with most available under an open license. The largest model currently available on the open market is Qwen3-32B, which has outperformed OpenAI’s o1 in a series of coding tests. However, the version with 235 billion parameters has yet to be released.
According to analysts, despite existing export restrictions, Chinese artificial intelligence developers are increasingly competing in the international market, approaching the level of Western leaders. Recently, OpenAI also introduced new AI models — o3 and o4-mini, which promise significant advancements in the development of “smart” models.
Alibaba stated: “Introducing Qwen3! We release and open-weight Qwen3, our latest large language models, including 2 MoE models and 6 dense models, ranging from 0.6B to 235B.”
Thus, Alibaba continues to strengthen its position in the artificial intelligence industry by offering new solutions that meet the modern demands for innovation and efficiency.