@inproceedings{ge-etal-2026-one,
title = "One Instruction Is a Benchmark: End-to-End Instruction-Following Evaluation with {F}lex{B}ench",
author = "Ge, Yuchen and
Sun, Yizhao and
He, Wei and
Ou, Jinpeng and
Niu, Wenzhe and
Shi, Chengshun and
Fan, Huichuan and
Ma, Biao and
Huang, Shuo and
Tian, Xin and
Gao, Jiuchong and
Hao, Jinghua and
He, Renqing",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1521/",
pages = "34726--34745",
ISBN = "979-8-89176-459-0",
abstract = "Real-world deployments of large language models (LLMs) increasingly rely on user-provided instructions that are long, precise, and tailored to specific workflows. In such settings, aggregate leaderboard performance is often a weak proxy for a model{'}s ability to faithfully execute the particular instruction that a practitioner intends to deploy. To address this gap, we introduce FlexBench/FlexEval, an instruction-conditioned evaluation framework that instantiates a specialized test suite from a user-provided target instruction. Given a target instruction, FlexBench (i) derives a set of verifiable evaluation dimensions and (ii) generates an instruction-conditioned conversation corpus. FlexEval then aggregates dimension-level judgments into two complementary metrics: Coverage, which measures workflow progress, and Achievement, which measures conditional correctness. We evaluate the framework on nine multi-turn target instructions spanning diverse scenario categories and workflow topologies, across nine contemporary LLMs over 3,888 generated dialogues. The results show that model rankings and failure modes vary substantially across target instructions, highlighting the need for instruction-conditioned evaluation beyond fixed aggregate benchmarks. Our source code is publicly available at https://github.com/Coryyy/FlexBench."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ge-etal-2026-one">
<titleInfo>
<title>One Instruction Is a Benchmark: End-to-End Instruction-Following Evaluation with FlexBench</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yuchen</namePart>
<namePart type="family">Ge</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yizhao</namePart>
<namePart type="family">Sun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wei</namePart>
<namePart type="family">He</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jinpeng</namePart>
<namePart type="family">Ou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wenzhe</namePart>
<namePart type="family">Niu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chengshun</namePart>
<namePart type="family">Shi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Huichuan</namePart>
<namePart type="family">Fan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Biao</namePart>
<namePart type="family">Ma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shuo</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xin</namePart>
<namePart type="family">Tian</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiuchong</namePart>
<namePart type="family">Gao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jinghua</namePart>
<namePart type="family">Hao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Renqing</namePart>
<namePart type="family">He</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Real-world deployments of large language models (LLMs) increasingly rely on user-provided instructions that are long, precise, and tailored to specific workflows. In such settings, aggregate leaderboard performance is often a weak proxy for a model’s ability to faithfully execute the particular instruction that a practitioner intends to deploy. To address this gap, we introduce FlexBench/FlexEval, an instruction-conditioned evaluation framework that instantiates a specialized test suite from a user-provided target instruction. Given a target instruction, FlexBench (i) derives a set of verifiable evaluation dimensions and (ii) generates an instruction-conditioned conversation corpus. FlexEval then aggregates dimension-level judgments into two complementary metrics: Coverage, which measures workflow progress, and Achievement, which measures conditional correctness. We evaluate the framework on nine multi-turn target instructions spanning diverse scenario categories and workflow topologies, across nine contemporary LLMs over 3,888 generated dialogues. The results show that model rankings and failure modes vary substantially across target instructions, highlighting the need for instruction-conditioned evaluation beyond fixed aggregate benchmarks. Our source code is publicly available at https://github.com/Coryyy/FlexBench.</abstract>
<identifier type="citekey">ge-etal-2026-one</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1521/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>34726</start>
<end>34745</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T One Instruction Is a Benchmark: End-to-End Instruction-Following Evaluation with FlexBench
%A Ge, Yuchen
%A Sun, Yizhao
%A He, Wei
%A Ou, Jinpeng
%A Niu, Wenzhe
%A Shi, Chengshun
%A Fan, Huichuan
%A Ma, Biao
%A Huang, Shuo
%A Tian, Xin
%A Gao, Jiuchong
%A Hao, Jinghua
%A He, Renqing
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F ge-etal-2026-one
%X Real-world deployments of large language models (LLMs) increasingly rely on user-provided instructions that are long, precise, and tailored to specific workflows. In such settings, aggregate leaderboard performance is often a weak proxy for a model’s ability to faithfully execute the particular instruction that a practitioner intends to deploy. To address this gap, we introduce FlexBench/FlexEval, an instruction-conditioned evaluation framework that instantiates a specialized test suite from a user-provided target instruction. Given a target instruction, FlexBench (i) derives a set of verifiable evaluation dimensions and (ii) generates an instruction-conditioned conversation corpus. FlexEval then aggregates dimension-level judgments into two complementary metrics: Coverage, which measures workflow progress, and Achievement, which measures conditional correctness. We evaluate the framework on nine multi-turn target instructions spanning diverse scenario categories and workflow topologies, across nine contemporary LLMs over 3,888 generated dialogues. The results show that model rankings and failure modes vary substantially across target instructions, highlighting the need for instruction-conditioned evaluation beyond fixed aggregate benchmarks. Our source code is publicly available at https://github.com/Coryyy/FlexBench.
%U https://aclanthology.org/2026.emnlp-main.1521/
%P 34726-34745
Markdown (Informal)
[One Instruction Is a Benchmark: End-to-End Instruction-Following Evaluation with FlexBench](https://aclanthology.org/2026.emnlp-main.1521/) (Ge et al., EMNLP 2026)
ACL
- Yuchen Ge, Yizhao Sun, Wei He, Jinpeng Ou, Wenzhe Niu, Chengshun Shi, Huichuan Fan, Biao Ma, Shuo Huang, Xin Tian, Jiuchong Gao, Jinghua Hao, and Renqing He. 2026. One Instruction Is a Benchmark: End-to-End Instruction-Following Evaluation with FlexBench. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 34726–34745, Budapest, Hungary. Association for Computational Linguistics.