DaisyRose 27B is an open-weight multimodal reasoning model from RadixArk, trained from Qwen3.8 27BOpens in new tab using reinforcement learning only on long-horizon agent tasks, with no supervised fine-tuning or distillation. It is aimed at agentic coding, multi-step reasoning, and instruction following.
It is a 27.8B-parameter dense model that accepts text and image input, returns text, and has a 262K-token context window. Thinking is on by default and can be disabled. It keeps the base model's architecture, tokenizer, and chat template, and is released under the Apache 2.0 license.
模态
输入/输出价格
25% 优惠
US$0.15 / US$1.875每 1M
上下文
262K
发布时间
2026年11月21日
目前无法通过 EU 区域内路由使用。 区域内路由文档
