Source: www.marktechpost.com/2026/11/18/hugging-face-qwen3-8-27b-pi-adapted-lora
Hugging Face has published Qwen3.8-27B-pi-adapted, a LoRA adapter for the Qwen3.8-27B model. The adapter is released under the Apache-2.0 license and was trained with reinforcement learning for the pi coding agent.
On SWE-bench Pro hard-51, the adapter's pass@1 score rises from 39.2% to 64.7%. On Terminal-Bench 2.1, it moves from 39.5% to 55.8%, according to the MarkTechPost report.
The report covers only these two benchmark results and the release itself. It does not describe the training setup in further detail.
Why it matters if you build agents
Developers running pi on Qwen3.8-27B can test the adapter directly, since it is openly licensed under Apache-2.0. The reported gains come from two benchmarks, so teams should run the adapter on their own coding and terminal tasks before changing their agent stack.
Get stories like this every weekday
One short email with the day's agent tooling news.