Key Researcher Steps Down from Alibaba's Qwen AI Project
- Junyang Lin, technical lead of Alibaba's Qwen AI project, announced his departure on March 3, 2026, via a social media post.[1]
- Qwen models have achieved over 600 million downloads and more than 170,000 derivative models on Hugging Face.[1]
- A fellow Qwen researcher suggested Lin's exit was involuntary, noting his recent involvement in launching Qwen3.5 small model.[1]
- Lin contributed to prior Alibaba projects including M6, OFA, and Chinese-CLIP.[1]
- Another Qwen researcher, Kaixin Li, also announced departure around the same time.[1]
Junyang Lin, the technical lead and public face of Alibaba's Qwen open-source AI project, announced his departure on March 3, 2026, in a brief social media post stating 'me stepping down. bye my beloved qwen.'[1] The move comes shortly after the launch of the Qwen3.5 small model and amid reports of potential internal changes at the Chinese tech giant's AI division.
Lin's Role and Qwen's Achievements
Lin served as the tech lead for Qwen, which Alibaba launched in beta in April 2023 and publicly released in September of that year. Under his leadership, the project expanded to include language models, vision-language models like Qwen-VL, audio models, math-specialized models, coding models, and the reasoning-focused QwQ series. By the Qwen3 release in April 2025, the models recorded over 600 million downloads and inspired more than 170,000 derivative models on Hugging Face, exceeding Meta's Llama in that metric.[1] Fortune magazine listed Alibaba on its 2025 Change the World List for its open-source AI efforts.[1]
Circumstances of Departure
Lin's announcement followed his participation in launching the Qwen3.5 small model just a day prior, where he thanked Elon Musk for praising its 'impressive intelligence density.'[1] Qwen contributor Chen Chang posted, 'I’m truly heartbroken. I know leaving wasn’t your choice. Just last night, we were side by side launching the Qwen3.5 small model. I honestly can’t imagine Qwen without you,' hinting at an involuntary exit.[1] Separately, Qwen researcher Kaixin Li announced, 'Signing off from @Alibaba_Qwen. Grateful for the chance to work with such brilliant minds. Proud of our impact. Onwards and upwards!'[1]
Lin's Background and Public Profile
Prior to Qwen, Lin contributed to Alibaba projects such as the mixture-of-experts model M6, the ICML 2022 paper OFA on multimodal pretraining, and Chinese-CLIP, a widely used Chinese-language vision-language model with over 2,000 GitHub stars.[1] He spoke candidly at a January 2026 Tsinghua University AI summit about China's compute disadvantages compared to the US, by one to two orders of magnitude, yet highlighted creative solutions like algorithm-hardware co-design.[1] Lin also delivered a keynote at ICLR 2025 on Qwen2.5's technical details.[1]
Recent Qwen Developments
Qwen3.5 features a 397 billion parameter open-weight model runnable on laptops with quantization, supporting 201 languages and multimodal inputs for text, image, and video.[2] It emphasizes agentic capabilities for multi-step tasks and integration with frameworks like OpenClaw.[2] The Qianwen App, powered by Qwen, integrates with Alibaba services like Taobao, Alipay, and Gaode for functions including ordering takeaways and booking flights, adding over 400 features.[3]
Companies mentioned
Alibaba Group Holding Limited, through its various subsidiaries, furnishes crucial technological infrastructure and extensive marketing channels. This comprehensive support allows merchants, brands, retailers, and other…