The USA–North America AI Olympiad is the selection pathway for U.S. students heading to the international AI olympiads. It runs a proctored written qualifier, an invitational two-day round hosted at Harvard and MIT, and a residential national camp where the team is chosen.
Registration for the 2027 season is open through 31 January 2027. Round 1 is on 19 February 2027.
How a USAAIO cycle runs, from registration to the contest itself.
Sign up through the Beaver Edge portal. Closes 23:59 EST.
Three hours, 12:00–15:00 EST, proctored in person at a USAAIO test centre.
Invitational, top Round 1 scorers only. Two in-person days — in 2026 these were hosted at Harvard and MIT.
Multi-day residential programme of team selection tests and lectures. The 2026 camp ran 8–12 June at Harvard.
What you are actually walking into, and what the rules allow.
A three-hour proctored paper taken in person. The 2026 set was 24 pages and nine problems: short theory questions on unsupervised learning, regularisation, bias–variance and F1, then linear algebra and PCA, matrix decomposition, derivatives, vectorised NumPy, token embeddings, CNN feature-map reasoning, custom PyTorch modules, and a constrained tabular classification task scored on macro-F1 against hidden test data.
Two in-person days for the top Round 1 scorers, hosted at major universities. Longer modelling problems replace the short-answer format, and the problems come with a published design rationale afterwards.
A residential week of lectures and team selection tests. Performance across the TSTs determines who is named to Team USA — eight students for IOAI and four for IAIO in the 2026 cycle.
PyTorch rather than TensorFlow, matching IOAI's requirement. Round 1 coding tasks are CPU-only unless a problem says otherwise, and individual problems impose their own import restrictions — the 2026 Problem 9 forbade deep networks entirely.
Built around the official USAAIO syllabus, not a generic ML course.
The mathematics and Python you need before any model makes sense: linear algebra, probability, calculus, and the NumPy/pandas toolchain.
Supervised and unsupervised learning with scikit-learn, plus the evaluation discipline that decides Round 1 scores.
Neural networks in PyTorch, from backpropagation by hand through CNNs, transformers, and generative models.
The topics that separate camp invitations from medals: regularisation theory, multimodal learning, and contest strategy.
USAAIO is a selection pathway. Doing well here puts you on a team headed somewhere.
An unofficial study guide, part of OpenIntelligence. Problem materials belong to USAAIO.
© 2026 OpenIntelligence.