RExBench: Best Coding Agents Achieve Only 33% Success in Autonomous Research Extension

RExBench, an ACL 2026 benchmark built on 12 real research papers, shows that even the strongest coding agents achieve only about 33% success in autonomously extending AI research, highlighting a significant gap toward fully autonomous scientific capability.

2026-10-16 ~ 2026-10-16 · 3 related posts

• RExBench: best coding agent completes only 33% of autonomous research extensions — najoungkim · 2026-10-16

2 near-duplicate retellings: najoungkim · najoungkim