SemComp-Bench

SemComp-Bench is a benchmark from ByteDance that evaluates whether AI models complete the actual task specified in a prompt, rather than merely generating plausible outputs. It shifts assessment focus to task completion fidelity, distinguishing it fr

Two-hop subgraph: this entity, every entity it directly relates to, and every entity those neighbors relate to. Drag a node, scroll to zoom, click to inspect — or click any neighbor and re-center the atlas there.