Acta Universitatis Sapientiae, Informatica Computer science

The Sapientia ECN AI Baseline Index: Benchmarking Large Language Models Against Student Performance in Competitive Programming

Sapientia Hungarian University of Transylvania, Zoltán Kátai, Dávid Iclănzan, Sapientia Hungarian University of Transylvania
Springer Nature · 2025-01-08 · DOI 10.47745/ausi-2024-0014

Abstract

We introduce the Sapientia ECN AI Baseline Index, a benchmark for evaluating the fundamental problem-solving capabilities of publicly available, stateof- the-art Large Language Models (LLMs) in competitive programming. Using basic prompting techniques on problems from the annual Sapientia Efficiency Challenge Networking (ECN) competition, we assess LLMs’ baseline performance, deliberately excluding more advanced enhancements like agentic systems or external knowledge retrieval. Our initial study compares LLM results with those of student teams from the ECN 2023 competition, analyzing both the number and types of problems solved, as well as score distributions. By providing a consistent, longitudinal measure, the ECN AI Baseline Index aims to track AI baseline capability advancement in complex problem-solving domains and offers insights into the evolving strengths and limitations of LLMs relative to peak and median student expertise.

Read at publisher →