LongBench v-2

'LongBench-v-2' is a benchmark used to evaluate the performance of models on tasks that require processing ultra-long contexts.

Papers using LongBench v-2 (3)

LongCrafter: Towards Diverse Long-Context Understanding via Evidence-Graph-Guided Instruction Synthesis2026Cache-Aware Prompt Compression:A Two-Tier Cost Model for LLM API Caching2026Evidence-State Rewards for Long-Context Reasoning2026