Models
RadixArk Releases DaisyRose-27B, an RL-Only Agentic Coding Model
RadixArk has released DaisyRose-27B, an open-weight 27.8B dense agentic coding model post-trained from Qwen3.8-27B using reinforcement learning only, without SFT or distillation, under Apache 2.0. It retains the base architecture with a 262,144-token context and single-GPU deployment options in BF16, FP8, NVFP4 and GGUF. RadixArk reports Terminal Bench 4.1 pass@1 rising from 5.6 to 12.0 and gains on coding benchmarks, while knowledge benchmarks improve only slightly. All benchmark figures are self-reported by RadixArk.
DaisyRose-27BRadixArkreinforcement learning