Qwen
4 articles in the last 30 days.
-
Introducing harness adapters: adapt your favorite model to your harness to improve performance and cost
RL-trained LoRA adapters per agent harness raise Qwen3.8-27B's Terminal-Bench 2.1 pass@1 under pi from 39% to 55%, while cutting average tokens per task from 2.9K to 1.5K. A reusable recipe is included.
-
DaisyRose-27B: Pushing the Frontier of Post-Training Efficiency
RRadixArk released DaisyRose-27B, an Apache 2.0 model built on Qwen3.8-27B via reinforcement learning. It raises Terminal-Bench 4.1 pass@1 from 5.6% to 12.0% and includes SGLang serving guidance.
-
Qwen4-Max vs. Claude Opus 5.5: The cheaper model costs you somewhere else.
Qwen4-Max is much cheaper per token than Claude Opus 5.5 but generates about twice the median output tokens. A hands-on test checks whether lower prices hold up in everyday work.
-
XYBench: Helping with the problem behind the question
XYBench is an 8,115-question benchmark testing whether AI assistants spot misconceptions in requests. All five models tested often answered the stated question instead of redirecting to the user's real goal.