<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="shen-etal-2026-vertical">
    <titleInfo>
        <title>Vertical Routing: A Cost-Efficient Collaboration Routing Framework</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Si</namePart>
        <namePart type="family">Shen</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Peijun</namePart>
        <namePart type="family">Shen</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Danhao</namePart>
        <namePart type="family">Zhu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <genre authority="bibutilsgt">journal article</genre>
    <relatedItem type="host">
        <titleInfo>
            <title>Transactions of the Association for Computational Linguistics</title>
        </titleInfo>
        <originInfo>
            <issuance>continuing</issuance>
            <publisher>MIT Press</publisher>
            <place>
                <placeTerm type="text">Cambridge, MA</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">periodical</genre>
        <genre authority="bibutilsgt">academic journal</genre>
    </relatedItem>
    <abstract>Routing Large and Small Language Models (LLMs SLMs) is commonly framed as query-level difficulty prediction, yet lightweight routers are often unreliable and stronger evaluators introduce non-trivial overhead. We propose VERTICAL ROUTING, a stage-level collaboration framework that avoids monolithic difficulty prediction by allocating the large model to critical subtasks and delegating the remaining generation to a small model. VERTICAL ROUTING instantiates two templates: (i) domain-specific templates that decompose a task into ordered stages with criticality scores, and (ii) a robust default template that follows a prefix-first prior for general queries. Under a token budget, we allocate large-model capacity to the most critical stages, or to a budgeted prefix under the default template. Experiments show that VERTICAL ROUTING outperforms the strongest baseline (RouterDC) by 2.0 points in the averaged metric, while significantly reducing token usage by 48.9% and large-model output share by 41.7%. Additionally, it enhances stability by lowering the standard deviation by 87.5%. These results highlight its advantages in accuracy, efficiency, and robustness.1</abstract>
    <identifier type="citekey">shen-etal-2026-vertical</identifier>
    <identifier type="doi">10.1162/tacl.a.802</identifier>
    <location>
        <url>https://aclanthology.org/2026.tacl-1.104/</url>
    </location>
    <part>
        <date>2026</date>
        <detail type="volume"><number>14</number></detail>
        <extent unit="page">
            <start>2301</start>
            <end>2317</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Journal Article
%T Vertical Routing: A Cost-Efficient Collaboration Routing Framework
%A Shen, Si
%A Shen, Peijun
%A Zhu, Danhao
%J Transactions of the Association for Computational Linguistics
%D 2026
%V 14
%I MIT Press
%C Cambridge, MA
%F shen-etal-2026-vertical
%X Routing Large and Small Language Models (LLMs SLMs) is commonly framed as query-level difficulty prediction, yet lightweight routers are often unreliable and stronger evaluators introduce non-trivial overhead. We propose VERTICAL ROUTING, a stage-level collaboration framework that avoids monolithic difficulty prediction by allocating the large model to critical subtasks and delegating the remaining generation to a small model. VERTICAL ROUTING instantiates two templates: (i) domain-specific templates that decompose a task into ordered stages with criticality scores, and (ii) a robust default template that follows a prefix-first prior for general queries. Under a token budget, we allocate large-model capacity to the most critical stages, or to a budgeted prefix under the default template. Experiments show that VERTICAL ROUTING outperforms the strongest baseline (RouterDC) by 2.0 points in the averaged metric, while significantly reducing token usage by 48.9% and large-model output share by 41.7%. Additionally, it enhances stability by lowering the standard deviation by 87.5%. These results highlight its advantages in accuracy, efficiency, and robustness.1
%R 10.1162/tacl.a.802
%U https://aclanthology.org/2026.tacl-1.104/
%U https://doi.org/10.1162/tacl.a.802
%P 2301-2317

Markdown (Informal)

[Vertical Routing: A Cost-Efficient Collaboration Routing Framework](https://aclanthology.org/2026.tacl-1.104/) (Shen et al., TACL 2026)

• Vertical Routing: A Cost-Efficient Collaboration Routing Framework (Shen et al., TACL 2026)

ACL

• Si Shen, Peijun Shen, and Danhao Zhu. 2026. Vertical Routing: A Cost-Efficient Collaboration Routing Framework. Transactions of the Association for Computational Linguistics, 14:2301–2317.