DaisyRose-27B
RadixArk's RL post-trained Qwen3.8 agent model: 1,200 curated long-horizon agent tasks over 18 days on 144 H200 GPUs; Terminal-Bench 4.1 pass@1 doubles from 5.6% to 12.0%.
Why it matters
A concrete, reproducible recipe showing a small lab can more than double an open model's agentic coding score with only 1,200 tasks and async RL — checkpoints in BF16, FP8 and NVFP4 under Apache 2.0.
What you could build with it
An indie developer could fine-tune a cheap open model with the same async-RL recipe on a few hundred of their own product's real terminal tasks, producing a specialist coding agent that outperforms the base model on their exact workflow at a fraction of frontier-model cost.
Does it hold up?
Too early to judge: the benchmark gains are the lab's own, but the training recipe and checkpoints are fully public, so independent replication can start immediately.
Built with DaisyRose-27B
- DaisyRose-27B Doubles Qwen3.8 Terminal-Bench Scoreemergent.sh · Coverage of the launch, the 1,200-task RL recipe, and the benchmark table.
- DaisyRose-27B: RadixArk Challenges the Limits of AI Post-Training With 57% Lower Training Costsundercodenews.com · Writeup of the async RL setup, the 37% MFU figure, and the benchmark gains.
Learn more
First spotted on hf: source.
More AI models
Edge0-145B: open 145B MoE that runs on a single consumer GPUEdge0 open-sourced 145B and 120B latent-MoE models derived from Qwen3.8-Flash-Next, plus an inference framework that…models · JEV 0.74Strata: run a 125B model on a gaming PCFree, MIT-licensed local inference engine that runs Qwen3.8-Flash-Next (125B MoE, about 6B active per token) on…models · JEV 0.73Qwen4-Max: Alibaba's new flagship model goes liveAlibaba's 2.4T-parameter flagship is now generally available via API, with a 1M-token context, native vision, and…models · JEV 0.67smolmaxxing: run 16B LLMs on a 1 GB TV box or a 2013 iPadAn MIT-licensed llama.cpp patch set that runs language models up to a 16B MoE on devices with about 1 GB of usable RAM…models · JEV 0.65Blue TTS v3: a 388MB open text-to-speech model with voice cloningAn open-source 388MB text-to-speech model (MIT) with voice cloning from about 10 seconds of speech, 7 languages, 15…models · JEV 0.65Haru-4BOpen full-duplex spoken-dialogue foundation model that listens and speaks simultaneously (7.5 Hz frames, 133 ms per…models · JEV 0.63
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.