RadixArk (Jason Chen, Hugging Face blog) · Agentic RL Model

DaisyRose-27B

DaisyRose-27B is RadixArk's first model, post-trained with reinforcement learning from Qwen3.8-27B using the Miles framework on about 1,200 long-horizon agent tasks. It raises Terminal-Bench 4.1 pass@1 from 5.6% to 12.0% and scores higher than the base model on every benchmark the post lists, including SWE-Bench Pro v2 (94.7 vs 91.3) and GPQA Diamond (90.7 vs 89.2). Checkpoints in BF16, FP8 and NVFP4 are released under Apache 2.0, and the model keeps the base's 256K context and image understanding.

Open the source ↗

Similar Language Models releases

  • Gemini 4 Neon

    Google DeepMind · Frontier LLM · Week 48, 2026

  • Gemini 4 Carbon

    Google DeepMind · Frontier agentic model · Week 47, 2026

  • Gemini 4 Hydrogen

    Google DeepMind · Agentic Coding Model · Week 47, 2026

  • LongCat-2.5

    Meituan (LongCat team) · Agentic MoE LLM · Week 47, 2026

  • Claude Fable 5.2

    Anthropic (announced via AWS Machine Learning blog) · Frontier Coding Model · Week 47, 2026

  • Inkling 2 Small

    Thinking Machines Lab · MoE agentic LLM · Week 46, 2026

Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.