ArXivMath

ArXivMath is a benchmark for hard reasoning in science, technology, engineering and math. The official description says: "Math reasoning benchmark over arXiv-style problems, reported in the Tencent Hy4 preview model card appendix". Below we describe in simple words what benchmarks of this type generally examine.

ArXivMath measures hard reasoning in science, technology, engineering and math. In detail, it checks:

• graduate- and expert-level questions in science and math
• multi-step derivations where each step must be correct
• problems written to defeat memorisation (novel or recent exam problems)
• correct use of formulas, units and logical inference

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.

Official methodology note: Math reasoning benchmark over arXiv-style problems, reported in the Tencent Hy4 preview model card appendix.

What is the ArXivMath benchmark?

ArXivMath is a benchmark for hard reasoning in science, technology, engineering and math. The official description says: 'Math reasoning benchmark over arXiv-style problems, reported in the Tencent Hy4 preview model card appendix'. Below we describe in simple words what benchmarks of this type generally examine.

What does ArXivMath test?

ArXivMath mainly measures hard reasoning in science, technology, engineering and math. Concretely, it checks: graduate- and expert-level questions in science and math; multi-step derivations where each step must be correct; problems written to defeat memorisation (novel or recent exam problems).

How is the ArXivMath score calculated?

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better. The declared maximum score of this benchmark is 100.0.

Where do I get ArXivMath data and scores?

The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/arxivmath/scores) or pull it directly from the Open LLM Radar API with any HTTP client.