OpenAI and Google LLC have announced that their latest reasoning models achieved gold medal-level performance in a recent programming competition.
The International Collegiate Programming Contest (ICPC) is the world’s most prestigious university-level programming competition, attracting participants from approximately 3,000 universities. OpenAI stated that its reasoning model achieved a perfect score, while Google’s algorithm solved 10 out of 12 problems presented during this year’s contest.
The ICPC finals took place on September 4, with 139 competing teams solving 12 problems provided by the organizers within a five-hour timeframe. If the submissions generated by AI from OpenAI and Google were completed by humans, they would have secured first and second place, respectively.
Both companies successfully solved problem C, which no human team answered correctly. The challenge involved calculating the most efficient way to fill a set of reservoirs. Google DeepMind researchers explained in a blog post today: “Problem C required finding a solution for distributing liquid through an interconnected network of pipes to a set of reservoirs, with the goal of identifying a configuration that allows all reservoirs to be filled as quickly as possible.”
Google participated using an “advanced” version of Gemini 2.5 Deep Think, a reasoning model launched in April. This algorithm tackles complex problems by generating a large number of potential answers in parallel, then refining and distilling them into a single response.
According to Google, the version of Gemini 2.5 Deep Think used in the ICPC employed AI agents to generate multiple potential solutions for each problem. These agents had access to a terminal, allowing them to run and test code. After generating initial code, they refined it to improve the quality of test responses.
Google researchers noted: “Reaching a gold medal level at ICPC has direct practical implications for software development. Beyond mathematics and coding, our achievement demonstrates a powerful new abstract reasoning capability. The skills required for ICPC—understanding complex problems, formulating multi-step logical plans, and implementing them flawlessly—are the same skills needed in many scientific and engineering fields.”
OpenAI participated in ICPC using GPT-5 and an experimental reasoning model that has not yet been publicly disclosed. GPT-5 solved 11 out of 12 problems with assistance from the reasoning model, which independently answered the final and most challenging problem in the set.
This milestone comes two months after models developed by Google and OpenAI won gold medals at the International Mathematical Olympiad. The tech giant revealed today that it used a version of Gemini 2.5 Deep Thinking in that competition.
