# Hugging Face — DaisyRose-27B: Pushing the Frontier of Post-Training Efficiency

- Company: Hugging Face (huggingface.co)
- Announced: 2026-11-18T19:54:56+00:00
- Category: new-model
- Coverage: 1 outlet
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/RadixArk/daisyrose-27b-pushing-the-frontier-of-post-trainin
- Record: https://forck.live/items/38200-daisyrose-27b-pushing-the-frontier-of-post-training-efficiency
- Subject: Platform
- Open weights: yes
- Models affected: DaisyRose-27B, Qwen3.8-27B, Qwen3.6-35B-A3B, GLM-5.3-Flash
- License: Apache 2.0
- Context window: 256K-token context window

RadixArk released DaisyRose-27B, its first model, built on Qwen3.8-27B and trained with reinforcement learning on long-horizon agent tasks. The company says the model raises Terminal-Bench 4.1 pass@1 from 5.6% to 12.0% and scores higher than the original Qwen3.8-27B on every benchmark it reports. Training used 1,200 curated tasks over 18 days on 144 H200 GPUs, with sessions up to 1M tokens. The checkpoints are available under Apache 2.0, and the model keeps Qwen3.8-27B's architecture, tokenizer, chat template, image understanding and 256K-token context window. RadixArk recommends serving it with SGLang and DSpark speculative decoding, and says its Miles backend cut training cost by 57% and doubled end-to-end throughput against an open-source Megatron-based trainer. The release run trained for weeks on 144 H200s, 63,000 GPU-hours in total, and its sandbox service ran about 944,000 sandboxes between mid-September and mid-November.

## Evidence

Verbatim from https://huggingface.co/blog/RadixArk/daisyrose-27b-pushing-the-frontier-of-post-trainin:

> Today we are releasing DaisyRose-27B , RadixArk’s first model, built on Qwen3.8-27B and trained with reinforcement learning using Miles . We trained it on long-horizon agent tasks with sessions up to 1M tokens, with just 1,200 curated tasks in 18 days on 144 H200 GPUs. DaisyRose-27B improves Terminal-Bench 4.1 pass@1 from 5.6% to 12.0% . Beyond Terminal-Bench 4.1, it also delivers gains on a variety of other reasoning, instruction following, and coding benchmarks.
> 
> The checkpoints are available under Apache 2.0 :
> 
> The model retains Qwen3.8-27B’s architecture, tokenizer, chat template, image understanding, and 256K-token context window . The longer training sessions use harness summarization to span multiple trajectories. See the model card for model properties and inference settings.
> 
> Run DaisyRose-27B
> 
> We recommend serving DaisyRose-27B with SGLang and DSpark speculative decoding. The SGLang cookbook provides launch configurations for different GPUs and checkpoint formats.
> 
> After starting the server with the served model name DaisyRose-27B on port 30000 , you can send a request using the OpenAI Python client:

## Around this story

Outlets this record can name and link:

- Extrapolator — DaisyRose-27B Training Achieves 37 Percent Model FLOPs: https://extrapolator.ai/2026/11/19/daisyrose-27b-training

---

Record: https://forck.live/items/38200-daisyrose-27b-pushing-the-frontier-of-post-training-efficiency
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md
