Tag
ReContext is a new training-free inference method that improves long-context reasoning by using model-internal relevance signals to build a query-conditioned evidence pool and replay it before final generation, achieving the best average rank on Qwen3 and Llama3 models across eight 128K-context datasets.
This paper introduces a four-condition diagnostic protocol to separate no-evidence answerability, oracle-evidence recoverability, full-context utilization, and retrieval-conditioned utilization in long-context and retrieval-augmented language models, tested on five open-weight models across multiple datasets.