RESEARCH · RESEARCH · #1684
Microsoft Research Asia open-sources Agent Lightning v1.0, a lightweight harnessed agentic RL framework
Microsoft Research Asia released Agent Lightning v1.0, a roughly 3,500-line open-source implementation of a Harnessed Agentic RL training paradigm that trains on the same agent harness used in deployment via an LLM proxy. The framework runs agents as native Kubernetes jobs, avoids reimplementing agent harnesses, and includes an end-to-end coding-agent example that raised Qwen3.5-9B’s Pass@1 on SWE-bench Verified from 41.8% to 56.4% using about 6,000 training samples.
KEY POINTS
- Microsoft Research Asia released Agent Lightning v1.0, a roughly 3,500-line open-source implementation of a Harnessed Agentic RL training paradigm that trains on the same agent harness used in deployment via an LLM proxy.
- The framework runs agents as native Kubernetes jobs, avoids reimplementing agent harnesses, and includes an end-to-end coding-agent example that raised Qwen3.5-9B’s Pass@1 on SWE-bench Verified from 41.8% to 56.4% using about 6,000 training samples.
- This matters because training directly on the deployed agent harness removes costly reimplementations, improves fidelity between training and deployment, and demonstrates data-efficient RL gains in a small, reproducible codebase.
WHY IT MATTERS
This matters because training directly on the deployed agent harness removes costly reimplementations, improves fidelity between training and deployment, and demonstrates data-efficient RL gains in a small, reproducible codebase.