AI Reports · Microsoft · 2026-10-07 · Product announcement
Agent Lightning v1.0: A 3,500-Line Lightweight Agentic RL Framework for Training Agents with Real Harnesses
Microsoft Research Asia introduced Harnessed Agentic RL, a training paradigm in which the same agent harness used in deployment takes part directly in reinforcement learning. Agent Lightning v1.0 provides a complete agent RL control plane in roughly 3,500 lines of code and runs natively on Kubernetes without paid commercial sandbox services. An end-to-end coding agent pipeline raised Qwen3.5-9B from 41.8% to 56.4% Pass@1 on SWE-bench Verified using about 6,000 training samples.