@inproceedings{garg-etal-2026-hydra,
title = "{H}y{DRA}: Hybrid Dynamic Routing Architecture for Heterogeneous {LLM} Pools",
author = "Garg, Aashna and
Roy, Siddharth Singha and
Fu, Shengyu and
Brancasi, Federico and
Jang, Jinu and
Cianci, Giuseppe",
editor = "Atanasova, Pepa and
Haffari, Gholamreza and
Wang, Minghan",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-industry.82/",
pages = "1377--1402",
ISBN = "979-8-89176-460-6",
abstract = "Production LLM deployments maintain heterogeneous model pools spanning order-of-magnitude cost differences, yet existing routers make binary strong-vs-weak decisions coupled to specific model identities, requiring retraining on every catalog change. We present HyDRA (Hybrid Dynamic Routing Architecture), which uses a lightweight ModernBERT encoder to predict capability requirements (reasoning, code generation, debugging, and tool use) for each query. It then selects the cheapest model meeting those requirements through shortfall matching. HyDRA decouples the learned predictor from the model catalog, enabling catalog changes without retraining. On SWE-Bench Verified, HyDRA is 0.3 QR points below always-strong (not statistically significant) at 54.1{\%} cost savings. On a held-out multilingual evaluation set, the same checkpoint retains 80.6{\%} of oracle-routing quality at 37.5{\%} cost savings, tunable to 79.1{\%} retention at 56.1{\%} savings. A 14-day A/B flight with close to 1M users per arm found statistically significant improvements in latency, time-to-first-token, reliability, and user activity. Deployed across GitHub Copilot VS Code Chat, GitHub Copilot CLI, and GitHub Copilot extensions, HyDRA demonstrates cross-lingual routing consistency across 16 languages and 4 language groups."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="garg-etal-2026-hydra">
<titleInfo>
<title>HyDRA: Hybrid Dynamic Routing Architecture for Heterogeneous LLM Pools</title>
</titleInfo>
<name type="personal">
<namePart type="given">Aashna</namePart>
<namePart type="family">Garg</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Siddharth</namePart>
<namePart type="given">Singha</namePart>
<namePart type="family">Roy</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shengyu</namePart>
<namePart type="family">Fu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Federico</namePart>
<namePart type="family">Brancasi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jinu</namePart>
<namePart type="family">Jang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Giuseppe</namePart>
<namePart type="family">Cianci</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: Industry Track</title>
</titleInfo>
<name type="personal">
<namePart type="given">Pepa</namePart>
<namePart type="family">Atanasova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gholamreza</namePart>
<namePart type="family">Haffari</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Minghan</namePart>
<namePart type="family">Wang</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">979-8-89176-460-6</identifier>
</relatedItem>
<abstract>Production LLM deployments maintain heterogeneous model pools spanning order-of-magnitude cost differences, yet existing routers make binary strong-vs-weak decisions coupled to specific model identities, requiring retraining on every catalog change. We present HyDRA (Hybrid Dynamic Routing Architecture), which uses a lightweight ModernBERT encoder to predict capability requirements (reasoning, code generation, debugging, and tool use) for each query. It then selects the cheapest model meeting those requirements through shortfall matching. HyDRA decouples the learned predictor from the model catalog, enabling catalog changes without retraining. On SWE-Bench Verified, HyDRA is 0.3 QR points below always-strong (not statistically significant) at 54.1% cost savings. On a held-out multilingual evaluation set, the same checkpoint retains 80.6% of oracle-routing quality at 37.5% cost savings, tunable to 79.1% retention at 56.1% savings. A 14-day A/B flight with close to 1M users per arm found statistically significant improvements in latency, time-to-first-token, reliability, and user activity. Deployed across GitHub Copilot VS Code Chat, GitHub Copilot CLI, and GitHub Copilot extensions, HyDRA demonstrates cross-lingual routing consistency across 16 languages and 4 language groups.</abstract>
<identifier type="citekey">garg-etal-2026-hydra</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-industry.82/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>1377</start>
<end>1402</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T HyDRA: Hybrid Dynamic Routing Architecture for Heterogeneous LLM Pools
%A Garg, Aashna
%A Roy, Siddharth Singha
%A Fu, Shengyu
%A Brancasi, Federico
%A Jang, Jinu
%A Cianci, Giuseppe
%Y Atanasova, Pepa
%Y Haffari, Gholamreza
%Y Wang, Minghan
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-460-6
%F garg-etal-2026-hydra
%X Production LLM deployments maintain heterogeneous model pools spanning order-of-magnitude cost differences, yet existing routers make binary strong-vs-weak decisions coupled to specific model identities, requiring retraining on every catalog change. We present HyDRA (Hybrid Dynamic Routing Architecture), which uses a lightweight ModernBERT encoder to predict capability requirements (reasoning, code generation, debugging, and tool use) for each query. It then selects the cheapest model meeting those requirements through shortfall matching. HyDRA decouples the learned predictor from the model catalog, enabling catalog changes without retraining. On SWE-Bench Verified, HyDRA is 0.3 QR points below always-strong (not statistically significant) at 54.1% cost savings. On a held-out multilingual evaluation set, the same checkpoint retains 80.6% of oracle-routing quality at 37.5% cost savings, tunable to 79.1% retention at 56.1% savings. A 14-day A/B flight with close to 1M users per arm found statistically significant improvements in latency, time-to-first-token, reliability, and user activity. Deployed across GitHub Copilot VS Code Chat, GitHub Copilot CLI, and GitHub Copilot extensions, HyDRA demonstrates cross-lingual routing consistency across 16 languages and 4 language groups.
%U https://aclanthology.org/2026.emnlp-industry.82/
%P 1377-1402
Markdown (Informal)
[HyDRA: Hybrid Dynamic Routing Architecture for Heterogeneous LLM Pools](https://aclanthology.org/2026.emnlp-industry.82/) (Garg et al., EMNLP 2026)
ACL
- Aashna Garg, Siddharth Singha Roy, Shengyu Fu, Federico Brancasi, Jinu Jang, and Giuseppe Cianci. 2026. HyDRA: Hybrid Dynamic Routing Architecture for Heterogeneous LLM Pools. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track, pages 1377–1402, Budapest, Hungary. Association for Computational Linguistics.