arXiv:2505. 19293v2 Announce Type: replace-cross Abstract: Long-context capability is considered one of the most important abilities of LLMs, as a truly long context-capable LLM enables users to effortlessly process many originally exhausting tasks -- e.
arXiv:2607. 09415v1 Announce Type: cross Abstract: Long-context processing has become increasingly important for large language models (LLMs), but simply extending the context window does not guarantee effective utilization of long inputs.
arXiv:2607. 08284v1 Announce Type: new Abstract: Large language models (LLMs) have demonstrated rapidly improving long-context capabilities, prompting a wave of benchmarks designed to evaluate them.
Large language models (LLMs) have demonstrated rapidly improving long-context capabilities, prompting a wave of benchmarks designed to evaluate them. However, existing long-context evaluations - from Needle-in-a-Haystack (NIAH) tests to more recent multi-hop reasoning and summarization tasks - predominantly measure average-case performance, and many are either saturated or lack robustness.
The paper introduces LongJudgeBench, a benchmark designed to evaluate large language models (LLMs) acting as judges for long-form text generation. It highlights that long-form evaluation requires complex, document-level assessments beyond simple length, such as organization, coverage, depth, consistency, and scenario-specific quality. Experiments show a significant reliability gap among current LLM judges, indicating instability across scenarios and limited effectiveness of rubrics or references.
The paper introduces LongHarness Bench, a new benchmark designed to evaluate both the effectiveness and efficiency of language model harnesses for long-context reasoning. It features tasks that require diverse retrieval strategies—such as lexical search and semantic matching—and strategic, adaptive reasoning over global and local context, with only a small subset of the context being useful at each step. Evaluations across multiple state‑of‑the‑art models and harnesses show that even strong combinations achieve only 68% macro‑average accuracy, and that the same model can vary markedly in efficiency depending on the harness used.
arXiv:2602. 15257v3 Announce Type: replace-cross Abstract: We present the first comprehensive, large-scale study of training long-context vision language models up to 344K context, targeting long-document visual question answering with measured transfer to long-context text.
arXiv:2507.06415v3 Announce Type: replace-cross Abstract: Long-context reasoning requires accurately identifying relevant information in extensive, noisy input contexts. In this work, we propose PERK...
arXiv:2607. 06160v1 Announce Type: cross Abstract: Synthesizing long-context supervised fine-tuning (SFT) data is a scalable way to enhance the long-context understanding of large language models (LLMs), yet existing approaches share three limitations: narrow task coverage, insufficient instruction difficulty, and a lack of faithfulness supervision.
arXiv:2505. 17315v2 Announce Type: replace Abstract: Recent language models exhibit strong reasoning capabilities, yet the influence of long-context capacity on reasoning remains underexplored.
arXiv:2606. 06203v1 Announce Type: cross Abstract: Input length and the position of relevant information are widely cited as the primary causes of degraded LLM long-context performance.
The paper presents a benchmark that compares seven long‑form generation frameworks across three granularities—single chapter, multi‑chapter, and whole book—using an anchor‑based LLM‑as‑a‑judge protocol to evaluate outlines directly. Results show no single framework dominates across all settings; performance depends on how well a framework’s output form matches the target granularity, with SuperWriter excelling in length‑constrained single‑chapter mode but losing advantage in whole‑book mode. The study finds only moderate correlation between outline and writing quality, supporting the idea that these two stages should be evaluated separately.
The paper introduces Highlight-Then-Summarize (H2S), a two-step approach that first highlights question-relevant evidence in long documents and then condenses it into a compact, question-conditioned summary before generating an answer. The authors built the H2S-Dataset with 6,647 examples spanning 11 benchmark families, and developed H2S-RL to reward evidence selection and summary construction. Evaluated on the H2S-Bench suite, the H2S-14B model outperforms larger open-source models, achieving the highest overall score and maintaining strong performance even with a reduced output budget.