RadixArk Releases DaisyRose-27B Agent-Focused Model
In short: RadixArk has launched DaisyRose-27B, a 27-billion parameter model built on Qwen3.8-27B, specifically fine-tuned for complex long-horizon agent tasks using reinforcement learning. The model demonstrates significant improvements across various reasoning and coding benchmarks, particularly in terminal-based automation.
This summary was generated automatically by AI from Hugging Face's publication. It is our own text, not a copy of the original — facts, figures and quotes belong to the source, linked above and below.
What changed?
- 1Significant gains in Terminal-Bench 4.1, doubling pass rate from 5.6% to 12.0%
- 2Improved performance on SWE Bench Pro v2 (94.7%) and GPQA Diamond (90.7%)
- 3Native reasoning output support via reasoning_content
- 4Available in BF16, FP8, and NVFP4 formats
| Parameter | Before | Now |
|---|---|---|
| Terminal-Bench 4.1 pass@1 | 5.6% | 12.0% |
| Context window | 256K | 256K |
Why it matters
This model represents a specialized advancement in agentic workflows where models must plan, reason, and execute interdependent steps over long sessions. For developers, the integration of native reasoning and improved instruction following offers a more reliable backbone for automation pipelines.
What it means for AI agents and contact centers
If your agents handle multi-step tasks in sandboxed environments, this model is worth testing for its superior handling of complex, long-horizon logic. Compare its reasoning latency and accuracy against current Qwen versions when deploying agentic workflows that require multiple tool calls.
🧪 Worth Testing
The model's specialized training for long-horizon agent tasks and its ability to return structured reasoning content make it highly relevant for complex, multi-step automation.
Sources
- Hugging FaceOfficialPrimary sourceOriginal article →„DaisyRose-27B: Pushing the Frontier of Post-Training Efficiency“18 Nov 2026, 21:54
- Published by source
- 18 Nov 2026, 21:54
- Found by our system
- 18 Nov 2026, 21:58
- Summary generated
- 19 Nov 2026, 20:05
This article was written by AI from the original source. Facts, numbers and prices come from the source; missing values are marked “Not specified”. Legal notice, copyright and privacy