UCCI: Calibrated Uncertainty for Cost-Optimal LLM Cascade Routing

Varun Kotte · 2026

Read UCCI: Calibrated Uncertainty for Cost-Optimal LLM Cascade Routing by Varun Kotte (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-2605-18796.txt.

Description

Varun Kotte offers UCCI, a calibration-first router for language model cascades that aims to cut cost without losing accuracy. Easy queries go to a small model and hard ones rise to a large model, but most routers use uncalibrated confidence and need retuning per workload. UCCI maps token margins to error chances and picks its threshold by cost minimization. On seventy-five thousand production queries, it cuts cost by about thirty-one percent while holding accuracy steady.

Catalog record

CREATORVarun Kotte
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE28133
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.18796

Browse the full library · About the Tower of Babel