UkisAI Swift Series / 27B, Flash Next and Bonsai 2 + GSQ-RCO / -63.4% thinking, x1.95 speed with xhigh accuracy

UkisAI has released Swift1.5 27B, an enhanced version of their previous model, which demonstrates improved agentic performance and reduced token usage. This new iteration uses 58.5% fewer thinking tokens while achieving a 0.35% higher score on Terminal Bench 2.1. It also avoids "overthinking error" loops, outperforming the base model. UkisAI encourages feedback and requests for their Swift models.