Lates News

Ali Qianwen announced the official release of the Qwen 3.5 medium model series, including Qwen3.5-Flash, Qwen3.5-35B-A3B, Qwen3.5-122B-A10B, and Qwen3.5-27B. It is introduced that the new models are more intelligent and require lower computing power. The performance of Qwen3.5-35B-A3B has surpassed Qwen3-235B-A22B-2507 and Qwen3-VL-235B-A22B, proving that architecture, data quality, and reinforcement learning are key to intelligent progress rather than simply stacking parameters. Qwen3.5-122B-A10B and 27B further narrow the gap between medium models and cutting-edge models, especially excelling in complex intelligent tasks. Qwen3.5-Flash is a hosted production version based on 35B-A3B, default supporting a context length of 1 million tokens with an embedded official toolkit.