Meta reported August 7 that its models earned gold-medal or equivalent performance across five international STEM Olympiad competitions: a gold medal at the International Mathematical Olympiad, gold-medal-level scores at the International Chemistry Olympiad and Romanian Masters of Mathematics, and perfect theory-exam scores at the Asian Physics Olympiad and International Physics Olympiad.
Tool use including search, coding, and calculators was disallowed to test pure reasoning, and the models used multi-agent reasoning throughout.