Microsoft releases Agent Lightning v1.0 agent RL framework
Understand the story
On October 8, 2026, Microsoft Research Asia released and open-sourced Agent Lightning v1.0, a reinforcement learning training framework for real harnessed agents. The team proposed the Harnessed Agentic RL training paradigm, whose core approach is to use an LLM proxy so that the agent harness actually deployed in production directly participates in reinforcement learning training, without needing to re-implement the agent within the training framework. The entire framework is about 3,500 lines of code. The framework is now open-sourced; this is the first report on the event, with no follow-up information yet.
AI overview of reporting · Updated 11 days ago
Reporting timeline
Explore different aspects of the story through its reporting.
- Microsoft ResearchSelectedAgent Lightning v1.0: A 3,500-Line Lightweight Agentic RL Framework for Training Agents with Real Harnesses
Microsoft Research Asia proposed the Harnessed Agentic RL training paradigm and open-sourced a rebuilt Agent Lightning v1.0, a roughly 3,500-line framework that lets the real agent harness used in deployment participate directly in reinforcement learning via an LLM proxy, with no need to reimplement the agent inside the training framework.