LiveCodeBench Pro: How Do Olympiad Medalists Judge LLMs in Competitive Programming?

Jingbo Shang (University of California, San Diego) · Peter Henderson (Princeton University) · Zihan Zheng (New York University) · Zerui Cheng (Princeton University) · Zeyu Shen (Princeton University) · Shang Zhou (University of California, San Diego) · Kaiyuan Liu (Department of Computer Science, University of Washington) · Hansen He (Canyon Crest Academy) · Dongruixuan Li (University of Waterloo) · Stanley Wei (Princeton University) · Hangyi Hao (McGill University) · Jianzhu Yao (Princeton University) · Peiyao Sheng (Sentient) · Zixuan Wang (Princeton University) · Wenhao Chai (Princeton University) · Aleksandra Korolova (Princeton University) · Sanjeev Arora (Princeton University) · Pramod Viswanath (University of Illinois, Urbana Champaign) · Saining Xie (New York University)
algorithmic categoriesalgorithmic contestsalgorithmic reasoningcompetitive programmingcomplex case analysisconfidently incorrect justificationsdata contaminationfine-grained diagnosticsimplementation-heavy problemslivecodebench promodel-generated submissionspass@1reasoning precisiontool augmentation

Recent reports claim that large language models (LLMs) now outperform elite humans in competitive programming. Drawing on knowledge from a group of medalists in international algorithmic contests, we revisit this claim, examining how LLMs differ from human experts and where limitations still remain. We introduce LiveCodeBench Pro, a benchmark composed of problems from Codeforces, ICPC, and IOI that are continuously updated to reduce the likelihood of data contamination. A team of Olympiad medalists annotates every problem for algorithmic categories and conducts a line-by-line analysis of failed model-generated submissions. Using this new data and benchmark, we find that frontier models still have significant limitations: without external tools, the best model achieves only 53\% pass@1 on medium-difficulty problems and 0\% on hard problems, domains where expert humans still excel. We also find that LLMs succeed at implementation-heavy problems but struggle with nuanced algorithmic reasoning and complex case analysis, often generating confidently incorrect justifications. High performance appears largely driven by implementation precision and tool augmentation, not superior reasoning. LiveCodeBench Pro thus highlights the significant gap to human grandmaster levels, while offering fine-grained diagnostics to steer future improvements in code-centric LLM reasoning.