To understand whether we're making genuine progress on reasoning, we entered our

Meta says its AI models reached gold-medal standard or better in five student science and mathematics olympiads while working without any tools. Its August 6 announcement lists perfect scores on the theory exams of the Asian Physics Olympiad and the International Physics Olympiad, a gold medal at the International Mathematical Olympiad, and gold-medal-level performance at the International Chemistry Olympiad and the Romanian Masters of Mathematics.

The no-tools condition is the point of the exercise as Meta frames it. The company says it entered the competitions to learn whether it is "making genuine progress on reasoning", and it disallowed search, coding and calculators. Every answer therefore had to come from the model's own reasoning, with no lookups and no computation handed off to software. Olympiad problems suit that kind of test because, in Meta's description, they demand deep chains of reasoning, creative insight and flawless argumentation.

The post's text leaves out most of what a reader would need to weigh the claim. It does not name the models, say who graded the answers or how, or explain why the mathematics olympiad result is a "gold medal" while the chemistry and Romanian results are described as "gold-medal-level performance". The physics claims are limited to the theory exams, as the post itself specifies. Until Meta names the models and publishes the grading details, these are the company's own reported results rather than independently checked ones.