Back to events
Models
Score 76Heat 13.5

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

Open SourceTrainingAgentsOpen WeightsReasoningDaisyRose-27BQwen3.8-27BRadixArk

Summary

RadixArk released DaisyRose-27B under the Apache 2.0 license, a model trained via reinforcement learning on long-horizon agent tasks using its open-source Miles framework. The model improves performance on Terminal-Bench and various reasoning benchmarks while lowering training costs.

Why it matters

It introduces open-weights reinforcement learning methodologies and efficiency gains tailored for long-horizon agent execution.

Editorial score

1 review
7
Significance
7
Novelty
9
Credibility
7
Reach
8
Actionability