AnalysisAI ModelsNovember 20, 2026

Qwen3.8-27B "super adapter" merges three agent harness LoRAs into one

Read original source →x.com

A single rank-96 LoRA stacking the Hermes Agent, OpenCode, and pi harness adapters lifts SWE-Bench Pro from 55% to 63% mean and cuts tool calls by up to 53–73%. The weight updates were nearly orthogonal, so concatenating them beat the base thinking model across harnesses.

How this story unfolded

2 days · 0 reports · 3 community posts · from Nov 18

  1. Nov 18
  2. Nov 20

More stories today

Open the live feed