OpenAI AI crushes all humans at top coding contest

At the AtCoder World Tour Finals 2026, an OpenAI AI agent entered an exhibition match against the world’s best competitive programmers—and won handily. The system solved all five problems in the Algorithm Division, including two exceptionally hard ones that stumped it for hours. No human competitor solved problems C or E. The runner-up, a human, scored about half the AI’s points. OpenAI’s Borys Minaiev admitted the result was surprising: problems D and E were tougher than any the team had tested. The AI uses a scaled test-time compute model comparable to GPT-5.6 and had no internet access. Six months ago, it couldn’t have solved most of these. Last year, an OpenAI model placed second at the same event; this time it took first. Why it matters: This marks a clear leap from AI being competitive to being dominant in abstract reasoning under time pressure—a domain once considered uniquely human.