IAIO splits the contest in two: a paper-style theoretical day of mathematical and algorithmic problems, then a practical coding day in Python on organiser-supplied machines. Its syllabus reaches into classical AI - search, logical reasoning, constraint satisfaction - that the other olympiads leave out.
IAIO 2026 was held in Ljubljana, Slovenia on 23–27 February 2026 with 102 students from 26 countries. IAIO 2027 runs in Vietnam, 25 February – 2 March 2027.
How a IAIO cycle runs, from registration to the contest itself.
National delegations register their interest through the official form before registration formally opens.
Member organisations run their own qualifiers - CAIO in Canada holds its National Qualifier in November.
Most delegations finalise their four-student team at a camp in the new year.
The first edition held in Southeast Asia.
What you are actually walking into, and what the rules allow.
Paper-style problems worked on a laptop, with answers typed up and submitted as PDF. The questions are mathematical and algorithmic: derive a result, trace an algorithm, justify why a method applies. This is closer to a maths olympiad paper than to a Kaggle notebook, and it is the round that most surprises students who prepared only by training models.
A supervised run at the exact machine and environment used for the practical round, so nobody loses contest time to a missing package or an unfamiliar editor.
Code-based model building, Python only. The organisers supply one laptop per competitor with the environment configured and datasets already loaded, and internet access is restricted.
Data preparation; supervised learning with regression, classification and regularisation; unsupervised learning with clustering and dimensionality reduction; reinforcement learning; AI search; logical reasoning; ML model evaluation; constraint satisfaction problems; kernel methods and SVM; and recommender systems. Note how much of that is pre-deep-learning material.
Final ranking is individual, with medals awarded on individual performance. An informal country ranking is published alongside, as at other olympiads.
Built around the official IAIO syllabus, not a generic ML course.
Data preparation, Python, and the probability and linear algebra the theory paper leans on.
Supervised, unsupervised, and kernel methods - the statistical half of the syllabus.
Search, logical reasoning, and constraint satisfaction. IAIO's distinguishing material, and the part most ML-trained students underestimate.
Reinforcement learning and recommender systems, where an agent or a ranking is the deliverable.
Writing a theory answer that scores, and building a practical solution offline under a time limit.
Students reach IAIO through their national olympiad.
An unofficial study guide, part of OpenIntelligence. Problem materials belong to IAIO.
© 2026 OpenIntelligence.