Announcement
Best-performing model reportedly resolves 109 of 244 open math problems in reuF2F benchmark
The creators say experts formalized the problems in Lean and built Leanimum-agent to compare models under one harness.
TLDR
The reuF2F creators say their benchmark contains 244 open math problems from Research Experiences for Undergraduates programs, formalized by experts in Lean. They built Leanimum-agent, a minimal coding agent, to compare models under the same harness. They report that even the best-performing model resolves only 109 problems.
Combined views
1.3K
5 Sources, first seen ago
