AutoRAGBench: AI agents build RAG systems that beat human designs

Tsinghua and Weibo released AutoRAGBench, the first benchmark testing whether LLM agents can build RAG systems from scratch in an open code space; GPT-6.1 scored 56.1, surpassing the best human-designed baseline.

2026-11-20 ~ 2026-11-22 · 2 related posts