iCreatAI News
Skip to content
Hugging Face Blog·· 2 days agoFeaturedAI score64

DaisyRose-27B: Pushing the Frontier of Post-Training Efficiency

DaisyRose-27B: Pushing the Frontier of Post-Training Efficiency

AI summary

RadixArk released DaisyRose-27B, its first model, built on Qwen3.8-27B and trained with reinforcement learning using Miles on long-horizon agent tasks with sessions up to 1M tokens, using 1,200 curated tasks over 18 days on 144 H200 GPUs. It raises Terminal-Bench 4.1 pass@1 from 5.6% to 12.0% and improves on every listed benchmark, with checkpoints available under Apache 2.0 in BF16, FP8 and NVFP4. The model keeps Qwen3.8-27B's architecture, tokenizer, chat template, image understanding and 256K-token context window.

Why it matters

The post details a concrete post-training recipe—task curation, an asynchronous Miles backend reporting 37% MFU, and a sandbox service—making it useful for teams weighing efficient agent RL training.

Source: Hugging Face Blog · huggingface.co