OJBench (language unspecified)
Code · 2025-06-19
OJBench is a competition-level code-generation benchmark of 232 problems from China's National Olympiad in Informatics (NOI) and the International Collegiate Programming Contest (ICPC), judged by hidden test cases. This row holds pass@1 results whose source does not state the solution language (C++ or Python), which changes scores materially, so it is a diagnostic; language-specific results use ojbench-cpp and ojbench-python. Higher is better.
Top models (higher is better)
| Model | Score |
|---|---|
| Kimi K2 Instruct | 27.1 |
| DeepSeek-V3-0324 | 24.0 |
| Opus 4 | 19.6 |
| GPT-4.1 | 19.5 |
| Sonnet 4 | 15.3 |
| Kimi Linear 48B A3B Instruct | 14.2 |
| Qwen3-235B-A22B | 11.3 |