Back to AI News
MarkTechPost

RadixArk Releases DaisyRose-27B, an RL-Only Agentic Coding Model

RadixArk has released DaisyRose-27B, an open-weight 27.8B dense agentic coding model post-trained from Qwen3.8-27B using reinforcement learning only, without SFT or distillation, under Apache 2.0. It retains the base architecture with a 262,144-token context and single-GPU deployment options in BF16, FP8, NVFP4 and GGUF. RadixArk reports Terminal Bench 4.1 pass@1 rising from 5.6 to 12.0 and gains on coding benchmarks, while knowledge benchmarks improve only slightly. All benchmark figures are self-reported by RadixArk.

AI Brief

Readers get a concrete look at a new open-weight RL-only coding model, including its training setup, deployment options and self-reported benchmark claims.

This week RadixArk released DaisyRose-27B, its first open-weight model. DaisyRose-27B is an agentic coding model from RadixArk, post-trained from Qwen3.8-27B with reinforcement learning (RL). The model card page states that it used RL only, with no SFT or distillation. It ships under Apache 2.0 as a drop-in replacement for the base model. The main result: Terminal Bench 4.1 pass@1 rises from 5.6 to 12.0. You can check the deployment options and specs here through our GPULeaderboard.com chrome extension.

TL;DR

  • Size: 27.8B dense parameters (0.46B vision encoder, 0.42B MTP head), 262,144-token context.
  • Runs on: 1 GPU. BF16 is about 52 GB, FP8 about 29 GB, NVFP4 about 23 GB (Blackwell only). Listed targets: H200, B200, GB300, RTX PRO 6000, RTX 5090, DGX Spark. GGUF also available.
  • Performance: Beats its Qwen3.8-27B base on all 7 reported benchmarks.
  • Best: IFBench 82.4, the top score among the 9 models RadixArk compared.
  • Worst: HLE (text only) 34.3, only +0.4 over the base and below Inkling 2 Small’s 44.1.
  • Bottom line: Best: Terminal Bench 4.1 more than doubles after 18 days of RL. Worst: it still trails much larger MoE models on hard agentic tasks.

What is DaisyRose-27B?

DaisyRose-27B keeps the Qwen3.8-27B architecture unchanged. The model card lists 64 decoder layers in a 3:1 pattern. Three Gated DeltaNet linear-attention layers alternate with 1 full-attention layer. Hidden size is 5,120. A 27-layer vision encoder handles images. One multi-token-prediction layer supports speculative decoding. Input is text and images, output is text, and it is a thinking model.

How was DaisyRose-27B trained?

RadixArk built about 1,200 curated tasks with Eigent AI, chosen from thousands of candidates. Each task targets a specific failure of the base model. Each sandbox attempt allows up to 1.5 hours, 100+ tool calls and 256k+ output tokens. Agent sessions reach up to 1M tokens, per the blog.

Training ran on Miles, RadixArk’s open-source RL framework, in fully asynchronous mode. The run used 144 H200 GPUs for 18 days. Of those, 128 handled agent inference and 16 handled training. Total compute was 63,000 GPU-hours.

RadixArk reports 37% MFU on its internal backend. An open-source Megatron-based trainer reached 16%. End-to-end throughput was 16.8k tokens/s versus 8.3k, a 2.0x speedup. RadixArk says this cut GPU time by 57%.

Sandboxes ran on AgentENV. RadixArk reports about 944,000 sandboxes from mid-September to mid-November. Peak concurrency hit 1,566. Creation success was 99% over 9 weeks. RadixArk puts the cost about 80% below popular cloud sandbox providers.

How does DaisyRose-27B perform on benchmarks?

All scores are self-reported by RadixArk. DaisyRose scores are the higher of its BF16 and FP8 runs.

BenchmarkDaisyRose-27BQwen3.8-27BChange
Terminal Bench 4.1 (pass@1)12.05.6+6.4
Terminal Bench 2.175.573.0+2.5
DeepSWE v1.146.242.2+4.0
SWE Bench Pro v294.791.3+3.4
HLE (text only)34.333.9+0.4
IFBench82.479.5+2.9
GPQA Diamond90.789.2+1.5

Source: RadixArk blog and model card.

The largest jump is Terminal Bench 4.1, which more than doubles. Coding gains are steady: +4.0 on DeepSWE v1.1 and +3.4 on SWE Bench Pro v2. Knowledge gains are small, at +0.4 on HLE and +1.5 on GPQA Diamond.

Against larger models, results are mixed. MiMo V2.6 Flash (309B total) leads Terminal Bench 4.1 at 16.1. Inkling 2 Small (594B total) leads DeepSWE at 72.1 and HLE at 44.1. DaisyRose-27B tops RadixArk’s 9-model table only on IFBench.

Interactive explainer · RadixArk

DaisyRose-27B: RL-only post-training on Qwen3.8-27B

How it is built, how it was trained, how it scores and how to run it. All figures from RadixArk's blog and model card.

27.8Bdense paramsIncludes a 0.46B vision encoder and a 0.42B multi-token-prediction head.
262,144token contextNative context per the model card. The blog cites agent sessions up to 1M tokens.
1 GPUsingle-card deployBF16 about 52 GB, FP8 about 29 GB, NVFP4 about 23 GB.
Apache 2.0licenseSame license as the Qwen3.8-27B base. Drop-in replacement.
64 decoder layers · 3:1 hybrid pattern (click to animate)
Gated DeltaNet (linear attention) × 48Full attention × 16Hidden size 5,120 · 27-layer vision encoder · 1 MTP layer
STEP 1

Curate tasks

~1,200 tasks from thousands of candidates

STEP 2

Agent rollouts

128 H200s in AgentENV sandboxes

STEP 3

Async RL update

16 H200s train with Miles

STEP 4

Ship weights

18 days, 63,000 GPU-hours

How do you deploy DaisyRose-27B?

Weights are on Hugging Face in BF16, FP8, NVFP4 and GGUF. NVFP4 requires a Blackwell GPU. RadixArk recommends SGLang with DSpark speculative decoding. That setup uses the Qwen3.8-27B-DSpark draft model. vLLM also works via vllm serve "RadixArk/DaisyRose-27B". Suggested thinking-mode sampling is temperature 1.0, top_p 0.95 and top_k 20.

How does DaisyRose-27B compare with similar-size models?

MetricsDaisyRose-27BQwen3.8-27BMuse Glimmer-30BQwen3.6-35B-A3B
DeveloperRadixArkQwen (Alibaba)Meta Superintelligence LabQwen (Alibaba)
Total params27.8B27B~29.6B35B
Active params27.8B (dense)27B (dense)~29.6B (dense)3B (MoE)
Context262,144262,144 (extensible to 1M)131,072+262,144 (extensible to 1,010,000)
LicenseApache 2.0Apache 2.0Apache 2.0Apache 2.0
InputText, imageText, image, videoText, imageText, image
Hardware1 GPU; 52 GB BF16, 23 GB NVFP4Not disclosed64 GB full precision; under 20 GB 4-bitNot disclosed
Terminal Bench 2.1*75.573.051.744.9
SWE Bench Pro v2*94.791.380.473.3
DeepSWE v1.1*46.242.24.40.0
HLE (text only)*34.333.922.022.2
IFBench*82.479.577.064.4
GPQA Diamond*90.789.283.586.0
SWE-bench Pro (official card)Not disclosed61.751.249.5

*From RadixArk’s comparison table, which cites each developer’s reports or leaderboards. SWE Bench Pro v2 figures trace to arXiv 2611.16289. The last row comes from each model’s own card.

Key Takeaways

  • RL only, no SFT: 1,200 tasks, 18 days, 144 H200s.
  • Terminal Bench 4.1 jumps from 5.6 to 12.0.
  • Fits 1 GPU: 52 GB in BF16, 23 GB in NVFP4.
  • Miles hit 37% MFU, 2.0x faster end to end.
  • Apache 2.0, but every score is self-reported.


Check out the model card, the technical blog and the Miles repo. You can check the deployment options and specs here through our GPULeaderboard.com chrome extension.

Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.

Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us

The post RadixArk Ships DaisyRose-27B, a 27B Agentic Coding Model Trained With Reinforcement Learning Only appeared first on MarkTechPost.

Be the first to rate this article.

Comments & Questions (0)

Captcha
Please be respectful — let's keep the conversation friendly.

No comments yet

Be the first to comment!