Challenging LLMs Beyond Information Retrieval: Reasoning Degradation with Long Context Windows
Open accessAbstract
As Large Language Models (LLMs) increasingly accommodate larger inputs, context windows spanning hundreds of thousands or even millions of tokens are touted as promising for a wide array of applications. However, the potential decay in reasoning ability for larger inputs may compromise their utility. This study introduces a new benchmark called Find the Origin, which progressively tests the efficacy of LLMs on a simple intellectual task as the size of the context window increases. The test, conducted on 14 different LLMs for comparative analysis, demonstrated that reasoning ability is dependent on input size. Additionally, three independent tests were performed with the GPT-4 Turbo model to demonstrate its reasoning degradation in different contexts as input size increases.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Long-context studies have shown that language models struggle to utilize relevant information within lengthy inputs, with performance often degrading well below the nominal context-window limit.
“Longcontext studies have shown that language models struggle to utilize relevant information within lengthy inputs, with performance often degrading well below the nominal context-window limit (Liu et al., 2024; Levy et al., 2024; Fraga, 2024; He et al., 2024; Hsieh et al., 2024; Tian et al., 2025).”
From Weighted Memory Tree: Remembering What Matters for Long-Horizon LLM Agents (2026) · cites this paper as Fraga, 2024
Cites
Levy, M., Jacoby, A., & Goldberg, Y. (2024) · Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2024.acl-long.818 · Cited by 119 · see citing works
Flagged first, then by citations
Cited by
Sorted by citations
Weighted Memory Tree: Remembering What Matters for Long-Horizon LLM Agents
Dao, Q., Kathalkar, P., & Eaton, K. (2026) · arXiv · doi:10.48550/arxiv.2608.20631
- Authors
- Fraga, N.
- Corresponding author
- Fraga, Natanael
- Venue
- Preprints.org
- Publisher
- MDPI AG
- Published
- 2024-08-22
- OpenAlex
- W4401831970
- Cited by
- 2
- References
- 1
- Citations by year
- 2026: 1 · 2025: 1
- Type
- posted content
- Language
- en
- Open access
- green
- doi.org
- Locations
- Preprints.org (acceptedVersion, cc-by)
- OSF Preprints (OSF Preprints) (submittedVersion, cc-by)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Multimodal Machine Learning Applications
- Keywords
- large language models, input size, GPT-4 Turbo, information retrieval
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
