Towards Resource-Efficient Multimodal Intelligence: Learned Routing among Specialized Expert Models

Mayank Saini, Arit Kumar Bishwas · 2025

Read Towards Resource-Efficient Multimodal Intelligence: Learned Routing among Specialized Expert Models by Mayank Saini, Arit Kumar Bishwas (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-06441.txt.

Description

Mayank Saini and Arit Kumar Bishwas route each incoming query to the cheapest model that can handle it well. Frontier language models now serve vision, audio, and document tasks, but running them on everything is too costly for real-time use. Small open models are cheaper yet stumble on complex or mixed-media questions.

A learned routing network reads each query and picks a fitting expert, balancing price against quality. Vision jobs pass through a two-stage open pipeline that revives classical methods where they still lead. On standard language and visual question benchmarks, the system matched or beat always-premium setups while cutting reliance on costly models by more than two thirds.

Catalog record

CREATORMayank Saini, Arit Kumar Bishwas
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE53 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-06441

Browse the full library · About the Tower of Babel