RadixArk open-sources DaisyRose-27B, lifting Terminal-Bench from 5.6% to 12% in 18 days
BanghuaZ · x · 2026-11-19
RadixArk releases its first open-weight model, DaisyRose-27B: RL-trained on Qwen3.8-27B with 1,200 curated long-horizon agent tasks, run for 18 days on 144 H200s. Using Miles and their own training backend, they stretched agent sessions up to 1M tokens, raising Terminal-Bench 4.1 from 5.6% to 12.0%. Full technical breakdown promised.
More from coding & agent
- Devs flag GPT's defensive programming for hiding bugs; OpenAI asks for /feedback cases — pvncher · 2026-11-19
- Flutter ships A2UI 1.0: agents respond with real interactive widgets, not chat text — rseroter · 2026-11-19
- Skills vs MCP vs RAG vs Memory: the four context mechanisms AI agents actually need — MaryamMiradi · 2026-11-19
- Anthropic details agent memory: 'dreaming' lets Claude organize its experience — daniel_mac8 · 2026-11-19
- The case for externalized memory as the real fix for continual learning — daniel_mac8 · 2026-11-19
- FT: US tech outages blamed on rush to fix bugs found by AI — nordicinst · 2026-11-19