GraphRAG-Bench/GraphRAG-Benchmark

GraphRAG-Bench addresses a critical gap in the RAG ecosystem: determining exactly where graph structures provide ROI over traditional vector-based RAG. With 390 stars and 49 forks, it has established a foothold as a credible research artifact, specifically linked to an ICLR 2026 submission. Its defensibility stems from the 'curation moat'—the specific datasets and task formulations required to test graph-traversal capabilities (multi-hop reasoning, global summarization). However, it faces stiff competition from industry-standard evaluation frameworks like RAGAS, which are rapidly expanding into graph-specific metrics, and from Microsoft's own GraphRAG implementation which includes internal evaluation suites. The project is currently a reference implementation for a paper; its long-term survival depends on whether it evolves into a living library (like MTEB) or remains a static research snapshot. The 'medium' frontier risk reflects that while labs like OpenAI and Anthropic are building better RAG, they rarely focus on building the evaluation benchmarks themselves, preferring to let the community define the standards they then optimize against.