Microsoft releases Agent Lightning v1.0 technical overview
The story so far
Microsoft Research released a technical overview on October 7, introducing the rebuilt Agent Lightning v1.0: the control surface code is about 3,500 lines, the agent runs in the form of Kubernetes jobs, and LLM agents enable existing harnesses to participate in reinforcement learning. Microsoft Research states that its Qwen3.5-9B encoding pipeline uses about 6,000 samples to increase SWE-bench Verified Pass@1 from 41.8% to 56.4%. The number is the result reported by the publisher.
AI-written from the reports · Updated 28 days ago
Coverage timeline
Follow the coverage to see every side of the story.
- Microsoft ResearchFeaturedAgent Lightning trains agents through their existing harness
Microsoft Research's October 7 technical overview describes the rebuilt Agent Lightning v1.0: roughly 3,500 lines of control-plane code, agents running as Kubernetes jobs, and an LLM proxy that lets an existing harness participate in reinforcement learning. Its reported Qwen3.5-9B coding pipeline improved SWE-bench Verified Pass@1 from 41.8% to 56.4% using about 6,000 samples.