RadixArk releases DaisyRose-27B for agent training
- Change DaisyRose-27B is built on Qwen3.8-27B and trained with reinforcement learning using the Miles framework.
- Detail RadixArk releases DaisyRose-27B as its first model.
- Detail DaisyRose-27B is built on Qwen3.8-27B and trained with reinforcement learning using Miles.
- Detail DaisyRose-27B improves Terminal-Bench 4.1 pass@1 from 5.6% to 12.0%.
- Detail RadixArk trained DaisyRose-27B on 1,200 curated tasks in 18 days using 144 H200 GPUs.
- Detail The training tasks are long jobs with several problems chained together.
- Detail RadixArk used Miles, an open-source reinforcement learning framework, to train DaisyRose-27B.
- Detail RadixArk is releasing BF16, FP8, and NVFP4 checkpoints of DaisyRose-27B for others to build on.
- Affected This affects builders who use AI models on reasoning, instruction following, and coding benchmarks.
Source: radixark.com ·