RadixArk open-sources DaisyRose-27B: reinforcement learning on Qwen3.8-27B with 1,200 long-horizon agent tasks lifts Terminal-Bench 4.1 from 5.6% to 12.0%
A full account of a small team post-training an open model for agent work on limited compute. For teams considering fine-tuning their own coding or operations agent model.
// Key points
- Built on Qwen3.8-27B with reinforcement learning using Miles; trained for 18 days on 144 H200 GPUs, with sessions up to 1M tokens, released under Apache 2.0.
- Terminal-Bench 4.1 pass@1 rises from 5.6% to 12.0%, and it scores higher than the original model on every benchmark listed.
- About 1,200 tasks were picked from thousands of candidates, each designed around a mistake the original model makes, with up to 1.5 hours, 100+ tool calls per attempt.
- It keeps the original architecture, tokenizer, chat template, image understanding and 256K context; SGLang with DSpark speculative decoding is the recommended serving setup.
Builder's takeAt 12% it is still far from closed frontier models, so I would not use it as a main model. What is worth studying is the task selection: only tasks the original model fails and cannot game, and 1,200 of them were enough to move the score. If you plan to post-train your own agent, build the task set from your failure cases like this first; it pays off more than piling up data.
// Background · from #Open models
Full timeline →- Nov 19 Cloudflare opens the weights of Clairic and Clairic-flash: 27B and 9B extraction models that read PDFs and bill only for input tokens
- Nov 19 Hugging Face shows Transformers streaming MoE experts from SSD: a 180B-parameter model runs at about 14 tok/s on a 32 GB MacBook
- Nov 19 AWS Lambda Managed Instances adds GPUs: run inference functions on EC2 G-series instances, with vLLM, SGLang and PyTorch base images