In the news
DaisyRose-27B: Pushing the Frontier of Post-Training Efficiency
Hugging Face · Published · 3 min read
In 30 seconds
- What happened
- RadixArk released DaisyRose-27B, a 27-billion-parameter model fine-tuned via reinforcement learning on long-horizon agent tasks, improving Terminal-Bench 4.1 from 5.6% to 12.0% pass rate.
- Why it matters
- Engineers building or deploying agentic systems should care, especially those working with reasoning tasks, software engineering benchmarks, or long-context agent training on limited compute budgets.
- Watch out
- The model was trained on only 1,200 curated tasks over 18 days; generalization beyond these specific domains and real-world agent performance remain unvalidated by independent evaluation.
- post-train
The patterns behind this
Each one covers how the technique works, when it earns its cost, and where it breaks.
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.