TSNBench: Benchmarking LLM Proficiency in Time-Sensitive Networking

Rubi Debnath, Daniel Bujosa Mateu, Luxi Zhao, Silviu S. Craciunas, Paul Pop, Sebastian Steinhorst · 2026

Read TSNBench: Benchmarking LLM Proficiency in Time-Sensitive Networking by Rubi Debnath, Daniel Bujosa Mateu, Luxi Zhao, Silviu S. Craciunas, Paul Pop, Sebastian Steinhorst (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-2605-09481.txt.

Description

Debnath and five colleagues present TSNBench, the first benchmark for judging large language models on time sensitive networking. This is the IEEE standards family that guarantees bounded latency for safety critical traffic. Such traffic serves vehicles, aviation, defense, and factories. The benchmark contains 939 expert validated multiple choice questions. It adds 100 open ended worst case delay computation tasks for two traffic shaping mechanisms. Ground truth comes from a verified network calculus solver and closed form bounds.

Sixteen models score between 67 and 95 percent on the multiple choice items. Yet they collapse on open ended delay computation. Even the best model misses by about 36 percent on one mechanism and about 42 percent on the other. Such errors dwarf real latency budgets. They could produce unsafe network settings. Multiple choice scores, the authors warn, flatter models in safety critical domains where exact calculation matters.

Catalog record

CREATORRubi Debnath, Daniel Bujosa Mateu, Luxi Zhao, Silviu S. Craciunas, Paul Pop, Sebastian Steinhorst
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE105 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.09481

Browse the full library · About the Tower of Babel