[KongchangAI]
ProductMini-SWE Agent

Mini-SWE Agent

A lightweight AI coding agent framework built by the SWE-Bench team, implementing a minimal agent run loop in approximately 100 lines of Python code. The framework achieves a solution rate of approximately 65% on the SWE-Bench Verified benchmark, demonstrating that effective software engineering agents can be constructed with a compact codebase.

Timeline (last 90 days)

Oct 8

The SWE Bench team built a streamlined framework called Mini-SWE Agent using approximately 100 lines of Python, and reported a solution rate of approximately 65% on SWE Bench Verified.

Unverified50%

All Facts (1)

Source Articles