From Tongyi Qianwen to Qwen
In April 2023, Alibaba Cloud offered corporate customers an invitation-only chatbot called Tongyi Qianwen. Named partly after the philosopher Mencius, the model was unveiled days later at the Alibaba Cloud Summit in Beijing and was said to be integrated into products such as DingTalk and Tmall Genie.
The open-weight era
In August 2023, Qwen-7B and Qwen-7B-Chat were released as open source, pretrained on over 2.2 trillion tokens. The same year brought the Qwen-VL vision-language branch and 72B and 1.8B sizes. In 2024, Qwen1.5, Qwen2 and Qwen2.5 rapidly expanded the open model lineup, with Apache 2.0 becoming the default license.
Reasoning and multimodality
In early 2025, models such as Qwen2.5-VL, QwQ-32B and Qwen2.5-Omni-7B added text, image, audio and video capabilities. In April 2025, Qwen3 launched with hybrid thinking and support for 119 languages, followed by Qwen3-Coder, Qwen-Image and Qwen3-Max.
2.4 trillion parameters
In August 2026, Qwen3.8-2.4T-A95B was released as open weights with 2.4 trillion parameters; companies with annual revenue above US$50 million require a custom license. Qwen3.8-27B was also offered under Apache 2.0. In September 2026, Alibaba said Qwen 4 is in training, with Qwen 4.5 and Qwen 5 projected at 5-10 trillion parameters.
