AgentSWE Benchmark: Coding Agents Fail to Build Agents

BAAI, CASIA and Renmin University released AgentSWE, a benchmark testing whether coding agents can build, edit and optimize agents from natural-language requirements. Existing agents scored poorly, averaging just 11.2 over 50 runs with none passing.

2026-10-13 ~ 2026-10-13 · 2 related posts

• AgentSWE benchmark: no coding agent comes close to delivering the agents it's asked to build — Jiahao Wang · 2026-10-13
• AgentSWE benchmark: coding agents score just 11.2/100 building the agents you actually want — _akhaliq · 2026-10-13