Harness Learning: 4B Model Rewrites Agent Code, Beats 35B Teacher

JHU and CMU researchers propose Harness Learning, where a 4B model trained with RL learns to rewrite agent harness code from failure reports, outperforming a 35B teacher model and generalizing to new tasks.

2026-10-04 ~ 2026-10-06 · 3 related posts

• CMU paper: RL-trained 4B proposer edits agent harness code, beats its 35B teacher — omarsar0 · 2026-10-04
• JHU & CMU Paper: Small Models Can Learn to Rewrite Agent Harnesses That Generalize to New Tasks — rohanpaul_ai · 2026-10-06
• JHU/CMU paper: a 4B model learns to rewrite agent harness code, beating its 35B teacher — rohanpaul_ai · 2026-10-06