LaunchAI ModelsNovember 17, 2026

Hugging Face LoRA lifts Qwen3.8-27B coding success to 64.7%

Read original source →huggingface.co

Hugging Face's harness-hub released Qwen3.8-27B-pi-adapted, a rank-32 LoRA trained with RL to drive the pi coding agent, scoring 64.7% on SWE-bench Pro (hard 51) vs 39.2% for the base model. Terminal-Bench 2.1 rose 39.5% to 55.8%, with thinking off matching base at max effort using ~4.5x fewer tokens.

How this story unfolded

2 days · 3 reports · 7 community posts · 10 of 12 shown

  1. Nov 17
  2. Nov 18
  3. Nov 19

More stories today

Open the live feed