Yang Fan
Algorithm Expert at Tongyi Lab, Alibaba Group.
Hangzhou, Zhejiang, China
I am an Algorithm Expert at Tongyi Lab, Alibaba Group, working on large language models as a member of the Qwen team. My work covers LLM training, fine-tuning, and data engineering. I have contributed to the Qwen series from the first Qwen release through Qwen3.5, Qwen3.8-Max, and the latest Qwen4-Max flagship, as well as multimodal models such as Qwen2-VL and Qwen2.5-Omni. See the Qwen blog for the latest releases.
Before joining Alibaba, I was a research intern at Microsoft Research Asia (2016–2022), mentored by Dr. Tao Qin. I received my Ph.D. in Computer Science from the University of Science and Technology of China in 2022.
My research interests span large language models, learning to teach (meta-learning frameworks that guide model training with a teacher model), neural architecture search, and neural machine translation. Our team won championship in 8 translation tasks at WMT19. See the publications page for the full list.
In my spare time I build hobby projects: Dreamulator, a science-grounded fantasy world simulator, and Oort, a toy programming language for exploring compiler design.
news
| Nov 16, 2026 | Qwen4-Max is released — the flagship of the new Qwen4 architecture: 2.4T-parameter MoE (71B activated), native vision, and a 1M-token context window. |
|---|---|
| Aug 02, 2026 | Qwen3.8-Max is released, and for the first time a Qwen-Max-class model gets open weights (2.4T-A95B). |
| Feb 16, 2026 | Qwen3.5 is released: natively multimodal, hybrid Gated DeltaNet + sparse MoE architecture, 201 languages. |
| May 14, 2025 | The Qwen3 technical report is out. See the Qwen blog for the latest releases. |
| Mar 26, 2025 | Qwen2.5-Omni, our end-to-end omni-modal model, is released. |