Can David Beat Goliath? On Multi-Hop Reasoning with Resource-Constrained Agents

Hojae Han, Heeyun Jung, Jongyoon Kim, Seung-won Hwang · 2026

Read Can David Beat Goliath? On Multi-Hop Reasoning with Resource-Constrained Agents by Hojae Han, Heeyun Jung, Jongyoon Kim, Seung-won Hwang (2026) 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-2601-21699.txt.

Description

Hojae Han, Heeyun Jung, Jongyoon Kim, and Seung-won Hwang ask whether small language model agents can handle multi-step reasoning on tight budgets. Reinforcement learning has empowered agents that search and use tools over many turns. The wins so far rest on heavy training with large models.

Under real resource limits, small agents fall into a weak regime. Few rollouts mean sparse exploration, sparse credit, and unstable learning. The authors study how such agents can still reason across hops. Their question is practical: most deployments cannot afford the high-cost regime, so reasoning must be made to work in the cheap one.

Catalog record

CREATORHojae Han, Heeyun Jung, Jongyoon Kim, Seung-won Hwang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE94 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-21699

Browse the full library · About the Tower of Babel