New benchmark tests whether context selection preserves evidence under token budgets
Researchers introduced CoRA-Bench, a benchmark evaluating whether context selectors that shorten long inputs during inference retain the evidence needed to answer a query.
Read the full story at arXiv cs.AI ↗
Timeline · 1 report
- 2026-11-10 05:00 · arXiv cs.AI
CoRA-Bench: Evaluating Evidence Preservation by Context Selectors Under Token Budgets