A benchmark of 200 recently resolved mathematical problems with formal Lean 4 statements is now available, enabling evaluation of LLMs' research-level problem-solving capabilities under controlled conditions. This development matters because it allows engineers running LLMs in production to assess and compare their models' ability to rediscover known solutions, a crucial step toward tackling currently open scientific problems. Evaluating LLMs on this benchmark can reveal whether they rely on memorization or genuine problem-solving capabilities.