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 stacks the near-orthogonal weight updates from Hermes Agent, OpenCode and pi, lifting SWE-Bench Pro from 55% to a 63% mean across harnesses. The merged adapter also cuts tool calls by 53–73% and beats the base thinking model.

2 sources

More stories today

Open the live feed