SCALE-QA is a new benchmark containing 3,000 audited questions designed to test if LLMs can maintain episode integrity in flat, unsegmented, multi-topic threads. It evaluates whether models can correctly infer dependencies from earlier conversation segments across 128k and 1M context windows.
HOW THIS AFFECTS YOU
●
builderThis helps you evaluate if your RAG or long-context implementation maintains thread coherence over long sessions.
●
researcherYou can use this to test if long-context improvements actually solve causal memory retrieval.