ukisai/Swift-Qwen3.8-27b

UkisAI releases Swift-Qwen3.8-27B, a fine-tuned derivative of Qwen3.8-27B that reduces thinking tokens by 58.3%. The model claims near-identical performance with a 1.95x speed-up on several tasks. It uses a transfer component from BottleCap AI. Independent verification of the efficiency gains is not provided in the source.