Announcement
RadixArk says DaisyRose-27B more than doubled its Terminal-Bench score in 18 days
RadixArk calls DaisyRose-27B its first open-weight model, built on Qwen3.8-27B and trained with reinforcement learning.
TLDR
Eighteen days, 144 H200 GPUs and 1,200 curated long-horizon agent tasks: that’s the training recipe RadixArk reports for DaisyRose-27B. The company says it stretched agent sessions to 1 million tokens and raised the Terminal-Bench 4.1 score from 5.6% to 12.0%.
Combined views
1.2K
7 Sources, first seen ago

Comments