Announcement
Qwen3.8-27B’s pi adapter cuts benchmark tool calls by 55–62%, its makers report
Harness-hub says its new LoRA adapter tunes Qwen3.8-27B specifically for the pi coding agent.
TLDR
A new LoRA adapter tunes Qwen3.8-27B for the pi coding agent. Its model card reports pass@1 rising from 29.4% to 62.7% on SWE-bench Pro’s hard-51 subset and from 39.5% to 55.8% on Terminal-Bench 2.1, with 55–62% fewer tool calls per task. One asterisk: the card says checkpoint selection used SWE-bench Pro scores and an 18-task Terminal-Bench subset.
Combined views
13.7K
2 Sources, first seen ago

Comments