The USA AI Olympiad (USAAIO) is rapidly becoming the most prestigious pre-college AI competition in the United States. It is for K–12 students in the U.S. and Canada. The competition tests both the theory and hands-on programming of machine learning — making it unlike any pure math or pure coding contest.
Top-performing students earn the right to represent Team USA at two landmark international Olympiads: the International Olympiad in Artificial Intelligence (IOAI) and the International Artificial Intelligence Olympiad (IAIO). USA teams have already earned Gold, Silver, and Bronze medals at both events.
“The AI Olympiad is neither a pure math olympiad nor a pure coding olympiad. To perform well, you need to excel in both mathematics and programming.”
— Official USAAIO Syllabus
That dual requirement is what makes USAAIO preparation both challenging and rewarding. A student who understands backpropagation only in code — without being able to derive the gradient by hand — will struggle with the theory questions. And a student who can prove theorems but cannot implement a neural network from scratch in PyTorch will struggle with the programming tasks. Both dimensions matter. Both dimensions matter, which is why a curriculum that balances theory and implementation in the right order (like the AlphaStar AI Program) tends to work better than self-studying scattered topics.
Preparing for USAAIO helps students:
The USAAIO runs on a three-stage pathway each academic year. Registration typically opens in fall, with Round 1 held in late winter.
Round 1 is a 3-hour proctored in-person exam in USAAIO test centers. Contestants answer both theory questions (written in Markdown with mathematical notation) and programming problems using Python, NumPy, and machine learning libraries. In the 2025 cycle, about 393 students participated nationally; roughly 76 (19%) qualified for Round 2. Award cutoffs for 2026 were 201/300 for Distinguished Honor Roll, 146/300 for High Honor Roll, and 98/300 for Honor Roll. The Round 2 qualifying cutoff was 98/300.
Round 2 is held in person across two days at major universities — in 2026, at Harvard and MIT. The format includes a more advanced competition, plenary talks by leading researchers, and an admissions information session. Finalists from this round are invited to the national camp.
The national camp is a multi-day program (in 2026, June 8–12 at Harvard) where team selection tests and lectures are held. The top performers are named to Team USA and represent the country at the IOAI and IAIO international Olympiads.
Python is the only programming language used in USAAIO. If you are new to programming, this is where to start. You need fluency with Python syntax, data structures (lists, dicts, sets), functions, classes, file I/O, and basic algorithmic thinking before any machine learning will make sense. Aim to feel comfortable writing programs of 50–150 lines without consulting references.
AlphaStar offers two sequential Python courses that require no prior programming background: CS21F-1 followed by CS21F-2. Together they build the programming foundation the rest of the AI track depends on.
Before diving into ML algorithms, build fluency with the scientific Python stack: NumPy for numerical computation, pandas for data manipulation, and matplotlib/seaborn for visualization. Learn to load real datasets, clean messy data, explore distributions, and communicate findings through charts. These libraries appear directly in USAAIO problems — you will be expected to use NumPy to implement algorithms from scratch, not just call high-level functions.
AlphaStar’s AI25F (Introduction to Data Science) teaches how to analyze data, find patterns, and communicate insights in Python — and includes a real project that can be geared toward regional STEM competitions. It assumes the Python courses and a pre-algebra math background.
This is the core of Round 1 preparation. You need to understand supervised learning algorithms — linear regression, logistic regression, SVMs, decision trees, k-nearest neighbors, ensemble methods — both how to derive them mathematically and how to apply them using scikit-learn. Key supporting concepts include the bias–variance tradeoff, cross-validation, loss functions, and regularization. On the unsupervised side, k-means clustering and PCA are explicitly on the syllabus; practice deriving the PCA eigenvalue problem on paper, then implementing PCA from scratch with NumPy. Probability and statistics matter too: Bayes’ theorem and Hoeffding’s inequality have appeared in theory questions.
AlphaStar’s AI31F (Introduction to Machine Learning) covers how core ML algorithms work and how to apply them in Python to real-world datasets. Projects from this level are competitive at top-tier STEM competitions such as ISEF and Broadcom MASTERS.
Deep learning is a major portion of the USAAIO, and the competition specifically requires PyTorch — not TensorFlow. Start with multi-layer perceptrons: understand the architecture, forward propagation, and backpropagation, and be able to compute gradients by hand for small networks. Learn essential components — affine layers, batch normalization, dropout — then move to CNNs for image tasks. On the generative side, the syllabus includes autoencoders, VAEs, GANs, UNet, and denoising diffusion probabilistic models, each of which requires both implementation skill and theoretical understanding.
AlphaStar’s AI41F (Introduction to Deep Learning) introduces neural networks and modern deep learning workflows in PyTorch, building intuition for how training works and exploring optimization techniques for stable learning.
The final and most rigorous stage focuses directly on USAAIO-format problems: timed, theory-heavy, and requiring both derivation and implementation under pressure. Transformers deserve special attention — the syllabus states contestants “shall have very solid and deep understandings of transformers,” covering attention mechanisms, the full architecture, and applications across NLP, vision transformers, and graph neural networks. This stage also revisits the advanced mathematical foundations (linear algebra, probability, convex optimization) that the theory questions draw on, and works through past USAAIO problems in a timed Google Colab environment.
AlphaStar’s two AI Olympiad courses (AI51F-1 and AI51F-2) focus on the theoretical grounds and math topics required for the USAAIO, with USAAIO-level problems solved in contest format. See the AI Olympiad course schedule →
Not sure which level to start at?
Get free consultation from AlphaStar AI experts — from first-time programmer to Olympiad contestant — and find where you fit.
The following topics are drawn from the official USAAIO syllabus. Every topic requires both theoretical understanding and programming ability — and each maps to one of the AlphaStar AI levels above.
📐 Mathematical Foundations
|
🐍 Core Programming Stack
|
🤖 Classical Machine Learning
|
🧠 Deep Learning (PyTorch)
|
🔮 Transformers & NLP
|
🎨 Computer Vision & Generative AI
|
