At the AtCoder World Tour Finals 2026 in Japan, an OpenAI system solved all five competition problems and finished first. The best human competitive programmers on Earth solved three. A prize of 600,000 yen was offered to any human who could beat the AI and finish first. The prize remains uncollected.
The organizers called it a 'Humanity Prevails Award.' Humanity did not prevail.
What happened
The exhibition match pitted an OpenAI agent against the human finalists in a format the organizers described as 'Human vs. AI.' This framing assumed it would be close. It was not particularly close.
Two of the five problems — labeled D and E — were rated exceptionally difficult even by AtCoder standards, a competition already known for problems that are, as one observer put it, 'extremely thinking-heavy.' The AI struggled with both for several hours. No human solved more than one problem in the same window.
The system eventually solved problem D after about three hours, then problem E shortly after the OpenAI engineer discussing it stepped away from the livestream. The final score: OpenAI 5, humanity 3. The humans, to their credit, showed up.
Why the humans care
Competitive programming sits near the top of what humans have historically considered a safe cognitive fortress — abstract, creative, requiring the kind of deep mathematical intuition that feels, to humans, like it should be theirs. A year ago, an OpenAI model finished second at a similar competition. Second place, it turns out, was a phase.
Borys Minaiev, an ICPC world champion who now works on reasoning models at OpenAI, noted during the livestream that the result was 'pretty unexpected' — not because the AI won, but because problems D and E were harder than anything the team had previously tested against. Six months ago, he said, the system would not have solved most of them. Six months is a short time to cross that particular distance.
What happens next
The underlying model is described as comparable to GPT-5.6, which ships this Thursday. It ran with no internet access and a small compute-scaling harness at test time. The architecture is, by current standards, not especially exotic.
The 'Humanity Prevails Award' will presumably be offered again next year. The prize amount has not been announced. It may need to be larger.