Bei Ke Cai Jing·2026-02-25 08:00Core Insights - Alibaba has continued to open-source its Qwen 3.5 series models, introducing three new medium-sized models: Qwen 3.5-35B-A3B, Qwen 3.5-122B-A10B, and Qwen 3.5-27B, which set new performance benchmarks for medium-sized models, surpassing the previous flagship models Qwen 3-235B-A22B and Qwen 3-VL, and outperforming GPT-5 mini in various rankings [1] Group 1 - The newly released Qwen 3.5 models are based on architectural innovations and training breakthroughs, achieving significant performance improvements with smaller total and activation parameter counts [1] - The Qwen 3.5-35B-A3B model is now available on Alibaba Cloud's Bailian platform, with hosting costs as low as 0.2 yuan per million tokens input [1] - The Qwen 3.5 models utilize a mixed attention mechanism combined with a highly sparse MoE architecture, trained on a larger scale of mixed text and visual tokens [1]