Research map / Large language models / Agents, code and math
LiveOIBench: Can Large Language Models Outperform Human Contestants in Informatics Olympiads?
Kaijian Zou, Aaron Xiong, Yunxiang Zhang, Xinliang Frederick Zhang, Yueqi Ren, Jirong Yang, Ayoung Lee, Shitanshu Bhushan, Lu Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use extended thinking instead of non-thinking mode (ablation) | Qwen3-32B · LiveOIBench | human percentile (%): 24.64 -> 42.00 | Improved |
| raise reasoning effort from standard to high (ablation) | GPT-OSS-120B · LiveOIBench | human percentile (%): 59.90 -> 72.88 | Improved |
| increase parallel samples (pass@k) (ablation) | LLM solvers · LiveOIBench | coding capacity | Mixed |
| increase sequential reasoning budget (ablation) | smaller models · LiveOIBench | coding performance | Improved |
| write solutions in C++ instead of Python (ablation) | GPT-OSS-120B · USACO (132 problems) | pass rate (%): 46.97 -> 59.09 | Improved |
| write solutions in Java instead of Python (ablation) | GPT-OSS-120B · USACO (132 problems) | pass rate (%): 46.97 -> 56.82 | Improved |
| rephrase task statements semantically (ablation) | GPT-OSS-120B, Seed-OSS, Qwen3-30B-A3B · LiveOIBench (71 problems) | relative score: small changes (e.g. -0.13) | No clear effect |
| mask 30-50% of statement tokens and remove examples (ablation) | GPT-OSS-120B, Seed-OSS, Qwen3-30B-A3B · LiveOIBench (71 problems) | relative score | Hurt |
| higher model familiarity with task statements (ablation) | GPT-OSS-120B · LiveOIBench | relative score: Pearson -0.02 (p 0.730) | No clear effect |
| higher familiarity with official solutions (ablation) | GPT-OSS-120B · LiveOIBench | relative score: Pearson -0.08 (p 0.151) | No clear effect |
Closest papers by meaning
- LiveCodeBench Pro: How Do Olympiad Medalists Judge LLMs in Competitive Programming? (NeurIPS 2025 · similarity 0.93)
- LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code (ICLR 2025 · similarity 0.90)
- LiveBench: A Challenging, Contamination-Limited LLM Benchmark (ICLR 2025 · similarity 0.88)
- CS-Bench: A Comprehensive Benchmark for Large Language Models towards Computer Science Mastery (ICLR 2025 · similarity 0.88)
- Benchmarking Large Language Models with Integer Sequence Generation Tasks (NeurIPS 2025 · similarity 0.88)
- WildBench: Benchmarking LLMs with Challenging Tasks from Real Users in the Wild (ICLR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
