About IMProofBench
Measures model performance on a private, regularly updated set of mathematical problems.
Key differences
Compare each service by the same aspect.
| Aspect | AI Proof Watch | IMProofBench |
|---|---|---|
| Subject coverage | Mathematics, theoretical computer science, and theoretical physics. | Mathematics. |
| Main purpose | Curate reported contributions and make claims, sources, and evidence easier to inspect; list linked human follow-up works. | Measures model performance on a private, regularly updated set of mathematical problems. |
| Assessment method | Four independent ratings cover AI involvement, technical verification, human understanding, and importance. Importance assumes the claim is correct; it is not a verification score. | Expert mathematicians grade solutions on a 0–3 scale; a score of 3 means a complete solution. The project also describes machine-checkable subproblems. |
| Email updates | Weekly updates across all topics, or selected-interest alerts that are daily by default and sent only when there is an update. | The leaderboard says the private problem set is regularly updated. We found no email subscription on the reviewed pages. |
When to use each
Use IMProofBench to compare model proof-generation performance on a controlled, contamination-resistant problem set.
By comparison, AI Proof Watch follows contributions reported in research sources across three fields. It records the claim and source context rather than scoring model attempts on a private benchmark, and its four ratings separate contribution, verification, understanding, and importance.
Sources
Feature comparison checked 20 October 2026. Public pages may change.