For questions about the benchmark or to discuss collaboration opportunities, please contact novamath.faq@gmail.com.
Nova-Math is a government-supported project (Korea’s Ministry of Science and ICT (MSIT) and the National Information Society Agency (NIA)) to build a calibrated mathematics benchmark that evaluates and challenges large language models (LLMs). We are recruiting problem writers to create original, high-quality items across four difficulty tiers: Challenge, Hard, Medium, and Easy. All accepted items are paid on a per-item basis, and authors will be included in a public paper describing the dataset, evaluation protocol, and baselines by year-end.
Tier | Mathematical Depth | Explanation | Compensation (per question) |
---|---|---|---|
Challenge | Graduate Research | Not solvable by current AI; typically only a few human experts in related areas can solve. A good Challenge problem has a novel core insight and resists back solving, guessing, or relying on heuristics. | $3,623 |
Hard |
Upper‑level undergrad or early graduate (e.g., Analysis, Complex Analysis, Algebra, Combinatorics, Probability, Topology) |
Comparable to the hardest problems at the International Mathematical Olympiad (IMO). One deep idea with a clean logical chain; little to no brute force casework. | $326 |
Medium |
Undergraduate (e.g., Analysis, Abstract Algebra, basic Graph Theory, Probability, intro Topology) |
Domestic Olympiad level tier. Crisp statement, single key idea, bounded computation. | $72 |
Easy |
High school or early undergrad (e.g., Calculus basics, Linear Algebra basics, Elementary Discrete Math, Elementary Probability) |
Middle/high school contest easy tier. Short solution, unambiguous statement. | $36 |
For questions about the benchmark or to discuss collaboration opportunities, please contact novamath.faq@gmail.com.