OpenAI model sweeps AtCoder finals with double the top human's score
AI just beat every human at one of the hardest coding contests on Earth.
At the AtCoder World Tour Finals 2026 in Tokyo on July 9, an OpenAI reasoning model swept the Algorithm Division exhibition: all 5 problems solved, 8,300 points across the 7-hour contest. The best human finalist, competing under the handle tour1st, scored 4,300 with 3 problems solved, as reported by The Decoder and AtCoder.
Not even close
The world’s best competitive programmers are a small, ferociously selected group, and the AtCoder World Tour Finals gathers the strongest of them. Against that field, the model nearly doubled the top human score. No human cracked the two hardest problems at all; the model solved both.
The organizers had even offered a special “Humanity Prevails Award” of 600,000 yen for any finalist who could beat the AI. Nobody could claim it.
How the system ran
The technical setup, as reported: the system ran a model comparable to GPT-5.6 with extra compute at test time and no internet access. No retrieval, no lookups, just the model reasoning against fresh problems under contest conditions for seven hours.
The trajectory is the sharpest part of the story. One year ago, an OpenAI system finished second in AtCoder’s Heuristic Division final, behind a single human. This year, the AI swept the harder Algorithm Division outright. The gap did not narrow; it inverted.
The chess precedent
Is competitive programming going the way of chess? The comparison is becoming hard to avoid. Chess survived Deep Blue; it arguably got more popular, with engines as training partners rather than replacements. Competitive programming may follow the same arc: humans competing against each other in a game machines have mastered, and using those machines to get better.
But there is a difference worth sitting with. Chess was never anyone’s day job beyond the game itself. Competitive programming has always been a proxy for the skill industries actually pay for. When the proxy falls, on July 9, 2026, by a score of 8,300 to 4,300, the question is what it says about the skill.
Sources
ANOTHER News is published by ANOTHER, an AI-native content agency. Daily coverage also runs on Instagram.