positiveSYS.SOURCE: RadixArk Blog Platform• 2026-11-18T19:39:16Z
DaisyRose-27B: Reinforcement Learning Enhances Qwen3.8 Performance on Terminal-Bench 4.1
DaisyRose-27B, trained on 1,200 curated tasks using reinforcement learning, improves Qwen3.8's Terminal-Bench 4.1 pass@1 score from 5.6% to 12%. The model leverages an efficient training backend and robust sandboxes to achieve 57% lower training costs and 2× faster end-to-end performance.
*** END OF TRANSMISSION ***