RadixArk Unveils DaisyRose-27B: Agentic LLM Boosts Reasoning Benchmarks
This summary and analysis were generated by AI from the original article at Hugging Face Blog and may contain errors (how Viqus works). Read the source for full details.
7
What is the Viqus Verdict?
We evaluate each news story based on its real impact versus its media hype to offer a clear and objective perspective.
AI Analysis:
While the hype surrounds the impressive benchmark gains, the real impact lies in the disclosed, efficient, and rigorous methodology for training complex agents.
Article Summary
RadixArk has launched DaisyRose-27B, their first model, which is built upon Qwen3.8-27B and further refined using reinforcement learning with Miles. The model was trained on a curated set of 1,200 challenging tasks involving long-horizon agentic work, sometimes spanning up to 1 million tokens. Key results show a marked improvement, raising the pass@1 score on Terminal-Bench 4.1 from 5.6% to 12.0%. The release emphasizes the rigorous training methodology, which involved creating tasks to specifically target the original model's weaknesses and utilizing an efficient, asynchronous training backend that significantly reduced compute costs compared to open-source alternatives.
Key Points
- DaisyRose-27B, based on Qwen3.8-27B, shows a substantial performance leap, notably increasing Terminal-Bench 4.1 pass@1 from 5.6% to 12.0%.
- The model's superior performance stems from training on complex, multi-step agent tasks in sandboxes, simulating real-world execution environments.
- RadixArk detailed its efficient training pipeline, claiming a 57% reduction in GPU time needed compared to open-source training frameworks.
Why It Matters
This release signals a continued industry push toward agentic capabilities, moving beyond simple prompt-response interactions to complex, multi-step problem-solving. The focus on long-horizon tasks and the detailed engineering of the training pipeline (task curation, sandbox execution) suggest that state-of-the-art LLMs are increasingly being defined by their ability to reliably plan and execute across extended workflows, which is a major commercial hurdle for current systems.

