IOI
The IOI is the world's most important programming competition for secondary school students. In the AI industry it serves as a tough benchmark for how well a model can think through difficult problems on its own and translate them into working code.
IOI stands for International Olympiad in Informatics. It is an annual competition in which students from around 90 countries compete against each other. Each country sends at most four young people, who have previously prevailed in national selection rounds. The tasks are not knowledge questions but difficult thinking problems that must be solved with a self-written program. Over two competition days, participants work on three tasks each within five hours. For a few years now, the IOI has also been appearing in reports about artificial intelligence, because companies have their programs measured against the same tasks.
Why companies advertise with IOI results
The IOI tasks are a good touchstone for a specific reason. They cannot be solved by memorization. Each task is newly invented and requires that a solution method be devised from scratch. This is considered the hardest part of what is called reasoning ability.
There is also a practical advantage: the grading is completely objective. A program is automatically fed many test cases and receives points for each case it passes. There is no room for subjective judgment, as there would be for an essay, for example. If an AI system scores points here, this can be verified and directly compared with the results of human participants.
That is why IOI results have become a marketing argument. When AI systems first achieved mid-range placements in 2024 and reported gold-medal level in 2025, this was one of the most-cited figures in the industry. Critics point out, however, that a competition result says little about how reliable a model is in everyday work.
What an IOI task requires
A typical task describes a problem in everyday language. For example: a country has cities and roads, and one must find the cheapest route under certain conditions. The obvious solution would be to simply try out all possibilities. But that fails precisely because it would take years for large inputs.
The real achievement lies in finding a clever method that delivers the same result in seconds. Such a method is called an algorithm: a precise set of instructions that leads step by step to the goal. Afterward, one must write the algorithm down error-free in a programming language, usually C++. Doing both under time pressure is the difficult part.
An AI model follows this same path in principle. It reads the task text, internally formulates a solution approach, and generates code from it. Modern systems are allowed to test their own code and revise it after finding errors. Unlike humans, they can also generate very many solution attempts in parallel and select the best one. Whether this advantage amounts to a fair comparison is debated among experts.
Where the term appears in the news
Most often one reads about the IOI in reports about new AI models. Providers such as OpenAI, Google DeepMind, or Anthropic like to cite scores from programming competitions when announcing releases. Alongside the IOI, the platform Codeforces is often mentioned, where programming tasks are solved year-round. Such figures are meant to show that a model does not merely formulate text but grasps problems deeply.
The IOI should not be confused with the IMO, the International Mathematical Olympiad. That competition involves proofs on paper, not runnable programs. Both are used in the industry as a benchmark and are sometimes mentioned in the same breath in press releases.
For young people themselves, the IOI is above all a career path. In Germany, the Bundeswettbewerb Informatik leads to the national selection. Former participants today often work in the research labs of exactly those companies that now measure their models against the same tasks.