@inproceedings{yang-etal-2025-100,
title = "100-{L}ong{B}ench: Are de facto Long-Context Benchmarks Literally Evaluating Long-Context Ability?",
author = "Yang, Van and
Jin, Hongye and
Zhong, Shaochen and
Jiang, Song and
Wang, Qifan and
Chaudhary, Vipin and
Han, Xiaotian",
editor = "Che, Wanxiang and
Nabende, Joyce and
Shutova, Ekaterina and
Pilehvar, Mohammad Taher",
booktitle = "Findings of the Association for Computational Linguistics: ACL 2025",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.findings-acl.903/",
doi = "10.18653/v1/2025.findings-acl.903",
pages = "17560--17576",
ISBN = "979-8-89176-256-5",
abstract = "Long-context capability is considered one of the most important abilities of LLMs, as a truly long context-capable LLM shall enable its users to effortlessly process many originally exhausting tasks {---} e.g., digesting a long-form document to find answers v.s., directly asking an LLM about it. However, existing real-task-based long-context evaluation benchmarks have a few major shortcomings. For instance, some Needle-in-a-Haystack-like benchmarks are too synthetic, and therefore do not represent the real world usage of LLMs. While some real-task-based benchmarks like LongBench avoid this problem, such benchmarks are often formed in a way where each data sample has a fixed sequence length, which not only makes them solely suitable for models with a certain range of context windows, but also lacks a proxy to know at what length the model/method-of-interest would fail. Last, most benchmarks tend to not provide proper metrics to separate long-context performance from the model{'}s baseline ability, so when conducting a cross-model/recipe comparison, such conflation makes the user unable to understand how exactly one model or recipe excels at the long-context task in relation to its baseline ability. To address these issues, we introduce a length-controllable, real-life reflective benchmark with a novel metric that disentangles baseline knowledge from long-context capabilities. Experiments demonstrate the superiority of our datasets in effectively evaluating LLMs. All assets are available at https://github.com/uservan/100-LongBench.git."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="yang-etal-2025-100">
<titleInfo>
<title>100-LongBench: Are de facto Long-Context Benchmarks Literally Evaluating Long-Context Ability?</title>
</titleInfo>
<name type="personal">
<namePart type="given">Van</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hongye</namePart>
<namePart type="family">Jin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shaochen</namePart>
<namePart type="family">Zhong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Song</namePart>
<namePart type="family">Jiang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qifan</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vipin</namePart>
<namePart type="family">Chaudhary</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaotian</namePart>
<namePart type="family">Han</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2025</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wanxiang</namePart>
<namePart type="family">Che</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joyce</namePart>
<namePart type="family">Nabende</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Shutova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohammad</namePart>
<namePart type="given">Taher</namePart>
<namePart type="family">Pilehvar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Vienna, Austria</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-256-5</identifier>
</relatedItem>
<abstract>Long-context capability is considered one of the most important abilities of LLMs, as a truly long context-capable LLM shall enable its users to effortlessly process many originally exhausting tasks — e.g., digesting a long-form document to find answers v.s., directly asking an LLM about it. However, existing real-task-based long-context evaluation benchmarks have a few major shortcomings. For instance, some Needle-in-a-Haystack-like benchmarks are too synthetic, and therefore do not represent the real world usage of LLMs. While some real-task-based benchmarks like LongBench avoid this problem, such benchmarks are often formed in a way where each data sample has a fixed sequence length, which not only makes them solely suitable for models with a certain range of context windows, but also lacks a proxy to know at what length the model/method-of-interest would fail. Last, most benchmarks tend to not provide proper metrics to separate long-context performance from the model’s baseline ability, so when conducting a cross-model/recipe comparison, such conflation makes the user unable to understand how exactly one model or recipe excels at the long-context task in relation to its baseline ability. To address these issues, we introduce a length-controllable, real-life reflective benchmark with a novel metric that disentangles baseline knowledge from long-context capabilities. Experiments demonstrate the superiority of our datasets in effectively evaluating LLMs. All assets are available at https://github.com/uservan/100-LongBench.git.</abstract>
<identifier type="citekey">yang-etal-2025-100</identifier>
<identifier type="doi">10.18653/v1/2025.findings-acl.903</identifier>
<location>
<url>https://aclanthology.org/2025.findings-acl.903/</url>
</location>
<part>
<date>2025-07</date>
<extent unit="page">
<start>17560</start>
<end>17576</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T 100-LongBench: Are de facto Long-Context Benchmarks Literally Evaluating Long-Context Ability?
%A Yang, Van
%A Jin, Hongye
%A Zhong, Shaochen
%A Jiang, Song
%A Wang, Qifan
%A Chaudhary, Vipin
%A Han, Xiaotian
%Y Che, Wanxiang
%Y Nabende, Joyce
%Y Shutova, Ekaterina
%Y Pilehvar, Mohammad Taher
%S Findings of the Association for Computational Linguistics: ACL 2025
%D 2025
%8 July
%I Association for Computational Linguistics
%C Vienna, Austria
%@ 979-8-89176-256-5
%F yang-etal-2025-100
%X Long-context capability is considered one of the most important abilities of LLMs, as a truly long context-capable LLM shall enable its users to effortlessly process many originally exhausting tasks — e.g., digesting a long-form document to find answers v.s., directly asking an LLM about it. However, existing real-task-based long-context evaluation benchmarks have a few major shortcomings. For instance, some Needle-in-a-Haystack-like benchmarks are too synthetic, and therefore do not represent the real world usage of LLMs. While some real-task-based benchmarks like LongBench avoid this problem, such benchmarks are often formed in a way where each data sample has a fixed sequence length, which not only makes them solely suitable for models with a certain range of context windows, but also lacks a proxy to know at what length the model/method-of-interest would fail. Last, most benchmarks tend to not provide proper metrics to separate long-context performance from the model’s baseline ability, so when conducting a cross-model/recipe comparison, such conflation makes the user unable to understand how exactly one model or recipe excels at the long-context task in relation to its baseline ability. To address these issues, we introduce a length-controllable, real-life reflective benchmark with a novel metric that disentangles baseline knowledge from long-context capabilities. Experiments demonstrate the superiority of our datasets in effectively evaluating LLMs. All assets are available at https://github.com/uservan/100-LongBench.git.
%R 10.18653/v1/2025.findings-acl.903
%U https://aclanthology.org/2025.findings-acl.903/
%U https://doi.org/10.18653/v1/2025.findings-acl.903
%P 17560-17576
Markdown (Informal)
[100-LongBench: Are de facto Long-Context Benchmarks Literally Evaluating Long-Context Ability?](https://aclanthology.org/2025.findings-acl.903/) (Yang et al., Findings 2025)
ACL