Alibaba Open Source Three Medium sized Qianwen 3.5 New Models, which can be directly deployed on consumer grade graphics cards On February 25th, following the open source of Qwen3.5-397B-A17B on New Year's Eve, Alibaba continued to open source the Qianwen 3.5 series models. This time, three medium-sized new models are open sourced, including Qwen3.5-35B-A3B, Qwen3.5-122B-A10B, and Qwen3.5-27B. Based on architectural innovation and training breakthroughs, the three open-source Qianwen 3.5 models have all set new performance records for medium-sized models, surpassing the larger flagship models Qwen3-235B-A22B and Qwen3-VL, and performing significantly better than GPT-5 mini on multiple charts. It is worth noting that the new Qianwen 3.5 model can even be directly deployed on consumer grade graphics cards, making it extremely developer friendly. At present, the hosting model Qwen3.5-Flash based on Qwen3.5-35B-A3B has been launched on Alibaba Cloud Bailian, with a minimum input of 0.2 yuan per million tokens. The Qianwen 3.5 model adopts a mixed attention mechanism, combined with a highly sparse MoE architecture innovation, and is trained on a larger scale of text and visual mixed tokens. The new model achieves performance improvement with a smaller total and activation parameter count. (New Beijing News)