The 30-second version. USAAIO runs an individual, three-round AI competition for U.S./Canadian secondary-school students under 20. Round 1 is open to everyone (proctored, done in Google Colab on CPU); the top ~1-in-5 reach Round 2; the best go to a June camp where team-selection tests narrow 12 finalists down to the 8 who represent the U.S. at the IOAI. It is deliberately both a math and a coding olympiad — you must derive and implement. Dates move every year, so confirm the current cycle on usaaio.org before anything else. Up next: the 2027 cycle — register by Jan 31 2027, Round 1 on Feb 19 2027.
What it is & who runs it
The USA AI Olympiad (USAAIO) is the organizing body that runs the USA–North America AI Olympiad (USA-NA-AIO), an artificial-intelligence competition for middle- and high-school students in the United States and Canada.
Its purpose is to serve as the national / North-American qualifier that selects Team USA for the international AI olympiads. This isn't just self-description: the international body's own site states that "the USA–North America AI Olympiad is organized by the USA AI Olympiad (USAAIO)," and names USAAIO (alongside Temple University) as the accredited U.S. organization, confirming that "Team USA is chosen from the top-performing students in the USA-North America AI Olympiad."
USAAIO — runs USA-NA-AIO for U.S. & Canadian students; fields the U.S. teams.
IOAI (the flagship), plus the sibling IAIO and IAI2O competitions.
Secondary-school students under 20 (marketed as "K-12"; binding rules say middle/high school).
Read the marketing carefully. USAAIO calls the U.S. a "founding country" and USA-NA-AIO the "unique / only pathway." Those phrases appear only on USAAIO's own site — IOAI and Wikipedia credit IOAI's founding to Bulgaria's LERAI Foundation (2024) and list the U.S. merely as "accredited." What is independently confirmed: USAAIO is the accredited U.S. organizer and its competition selects Team USA.
Who can compete
There are two eligibility tiers, and conflating them is the most common misconception. Anyone who meets the general bar can enter and compete through all three rounds — but the seat on the international IOAI delegation carries a stricter residency rule set by IOAI itself.
| Requirement | To enter & compete in USA-NA-AIO | To represent Team USA at the IOAI |
|---|---|---|
| Age | Under 20 in the competition year | Same — IOAI caps at under 20 |
| Enrollment | Full-time middle- or high-school student; not a full-time university student that year | Enrolled in a secondary school |
| Residency / citizenship | U.S. or Canadian citizen or permanent resident, or a full-time student studying in the U.S./Canada | U.S. citizen, U.S. permanent resident (green card), or an international student attending a U.S. secondary school |
Key practical consequence: Canadian students may compete in USA-NA-AIO, but under IOAI's rules they cannot take a Team USA slot at the IOAI. If the international team is your goal, confirm your residency status early. (A minor wording gap exists between two official pages on the age cutoff — "under 20 on the first day of the international competition" vs. "under 20 by August 30 of the competition year"; since the IOAI runs in late August these are roughly equivalent.)
Timeline & registration
Round 1 has drifted from late March (2025) to late January (2026) to mid-February (2027). Treat the table below as a pattern, not a promise, and always confirm the current cycle on usaaio.org. Registration typically opens the previous summer and closes in late January. You must secure a proctor before you register.
| Cycle | Registration | Round 1 | Round 2 | Camp / Round 3 |
|---|---|---|---|---|
| 2025 | — | Mar 24 | Apr 27 · MIT | Jun 9–13 · online (3 selection tests) |
| 2026 | Jul 1 2025 – Jan 26 2026 | Jan 30 2026 12–3pm ET | Apr 4–5 · Harvard & MIT | Jun 8–12 · Harvard |
| 2027Next | closes Jan 31 2027 11:59pm EST | Feb 19 2027 12–3pm EST | Spring 2027 (TBA) | Jun 2027 (TBA) |
Action for a prospective 2027 competitor: register before Jan 31, 2027 (11:59pm EST) and be ready for Round 1 on Feb 19, 2027 (12–3pm EST). Line up an approved proctor well ahead of the deadline — an unproctored score is not certified and cannot advance you.
Format & the three rounds
The competition is individual (not team-based) and runs three sequential, cumulative stages — you qualify for each from the previous one. (A widely repeated "two rounds" description is wrong; the organizer's own rules say three, and that claim was explicitly refuted in verification.)
Every round is taken onsite and proctored — a supervised, computer-based Colab exam, not an at-home test. Round 1 mixes typed theory (Markdown / math cells) with programming cells. Recent Round 1 problems have used multiple-choice and numeric-answer formats (per the Beaver-Edge prep forum — lower-confidence detail).
Round 1 is CPU-only — GPUs are "neither needed nor allowed," so your solutions must be efficient without acceleration. Round 2 is where GPU-scale work appears (an L4 GPU is provided). Practice hitting time limits on CPU.
Syllabus & topics tested
The USAAIO syllabus is broad and demands you can both derive the math and implement the method. The four pillars:
Mathematical foundations
Classical machine learning
Deep learning
Coding stack
Aiming for the IOAI team? The international syllabus is materially broader. The IOAI 2026 syllabus (revised yearly by its Scientific Committee) organizes topics into four areas and reaches well past USAAIO's Round-1 core:
- Foundational skills & classical ML — regression, SVM, PCA, clustering
- Neural networks & deep learning — attention, Transformers, diffusion models
- Computer vision — CLIP and modern architectures
- NLP & audio — BERT, pre-trained LLMs (open-source and API-based)
Notably, 59 of 60 IOAI subtopics are labeled "Practice" or "Both" — only one is theory-only. This is an implementation-first competition.
Difficulty & scoring
USAAIO is explicit that a strong contestant must excel at both theory and programming: "knowing theory without knowing how to program … is insufficient. Likewise, knowing how to program without understanding the theory … is also insufficient." That dual demand is the defining difficulty.
Round 1 is a ~3-hour proctored Colab exam interleaving theory answers (typeset in Markdown / math notation) with live programming problems — under time pressure, on CPU.
Advancement thresholds vary by year and aren't officially published. Reported data points: a Round 1 cutoff around 53/300 (2025) and 98/300 (2026), with roughly 19–20% advancing — in 2025, about 76 of 393 entrants.
IOAI & team selection
USAAIO is the accredited U.S. organizer for the International Olympiad in Artificial Intelligence (IOAI) — the flagship global competition for high-school students, first held in Burgas, Bulgaria in 2024 — and the presented U.S. route to two sibling events, the IAIO and IAI2O.
The selection funnel is concrete and verifiable: from Round 2, the top 12 advance to the June USAAIO Camp; there, intensive team-selection tests narrow the field to the 8 students — two teams of four — who form the U.S. delegation to the IOAI. At the international event, contest tasks span machine learning, computer vision, and NLP, and require full ML workflows: writing code, fitting models, and running inference.
Bottom line for planning: USAAIO Rounds 1–2 test a focused classical-ML + CNN core, but the camp and the IOAI itself reward the broader modern-DL toolkit (Transformers, diffusion, CLIP, BERT, LLMs, CV, NLP). If your ambition is the medal stage, prepare the wider IOAI syllabus in parallel — don't stop at the Round 1 material.
A preparation strategy
Build genuine dual competence
Study the four math areas (linear algebra, probability/stats, multivariable calculus, convex optimization) and implement every ML/DL method twice — once from scratch, once with libraries. Neither skill alone clears the bar.
Master the Python / Colab stack on CPU
Get fast and fluent with NumPy, pandas, scikit-learn, and PyTorch inside Google Colab. Because Round 1 is CPU-only, practice writing efficient solutions without GPU acceleration.
Sequence your study by difficulty
Start with classical ML — derive and implement PCA, regression, SVM, k-means, and ensembles. Then deep-learning foundations and CNNs. Then, if you're targeting the IOAI team, the modern-DL layer: Transformers, attention, diffusion, CLIP/BERT, pre-trained LLMs, computer vision, and NLP/audio.
Rehearse under timed, exam-like conditions
Work the official USAAIO past problems (2025 and 2026 Round 1 & Round 2 sets are posted, plus a 2026 "design & rationale" doc) inside a 3-hour Colab window to build speed and format familiarity.
Handle the logistics early
Secure an approved proctor before registering, mark the registration deadline, and — if you're aiming for the Team USA slot — confirm your citizenship/residency status against the IOAI rule up front.
Study resources
On paid prep programs. Third-party providers — AlphaStar Academy Comrite Learning Beaver-Edge forum — market USAAIO-specific courses and sample problems. They exist and can help, but they're commercial and unofficial, weren't quality-assessed here, and are no substitute for the official syllabus and past problems. Start free; pay only if you've hit a specific wall.
Caveats & open questions
- Dates & venues — Round 1 has ranged from January to March; Round 2 venues and GPU details are year-specific. Verify live.
- Scoring thresholds — advancement cutoffs change annually and aren't officially published.
- The IOAI syllabus — revised each year by the IOAI Scientific Committee, so track the current version.
- The exact current-cycle scoring rubric and per-problem point allocation.
- The camp's internal structure and the precise criteria that cut 12 finalists to 8.
- Registration/participation costs and specific approved-proctor requirements.
- Whether Round 2 is in-person at named universities this cycle, and how non-U.S.-eligible Canadians are handled beyond it.
Sources
Official / primary
Corroborating & third-party
- Institute of Competition Sciences
- Beaver-Edge prep forum
- IOAI — U.S. organizer page
- AlphaStar Academy, Comrite Learning, IvyMax — commercial prep (unofficial)
- Wikipedia — IOAI (background / founding)