WebChallenger web agent matches proprietary reasoning models at lower inference cost

WebChallenger uses a DOM-derived structured page representation called PageMem plus divide-and-conquer attention, persistent site memory, and procedural pattern caching to close the gap with expensive reasoning models on web navigation benchmarks. The architecture targets the repetitive-task use case where frontier model inference cost is prohibitive.