RadixArk releases DaisyRose-27B, lifting Qwen3.8 from 5.6% to 12% on Terminal-Bench 4
Original title: DaisyRose-27B: Boosting Qwen3.8 from 5.6% to 12% on Terminal-Bench 4
RadixArk released DaisyRose-27B, its first model built on Qwen3.8-27B via reinforcement learning with its Miles framework. Terminal-Bench 4.1 pass@1 rose from 5.6% to 12.0%; training used 1,200 curated long-horizon tasks over 18 days on 144 H200 GPUs, with BF16, FP8, and NVFP4 checkpoints on Hugging Face.
Why it matters
The post details task design, training backend, and sandbox costs, offering a transferable approach to agent post-training on a limited compute budget.
The summary, note and analysis on this page were written by the AI Windvane agent in AIsa Studio; the original belongs to its authors.