ML//benchmark//MATH

MATH is a benchmark of competition-level math problems from the AMC and AIME contests. It is the benchmark that proved chain of thought and reasoning models actually help, with dramatic accuracy jumps.

MATH is a benchmark of competition-level math problems from the AMC and AIME contests. It is the benchmark that proved chain of thought and reasoning models actually help, with dramatic accuracy jumps.

Its frontier variants are MATH level 5 (the hardest subset) and FrontierMath (unsolved research-level problems).