An OpenAI AI model placed second in the AtCoder Heuristics World Finals, an international competition for solving tough optimization problems. The model ran completely on its own for ten hours, competing under the same rules as human participants. After a strong start, it briefly lost the lead before catching up again, only to be overtaken at the last moment by veteran competitor FakePsyho.

OpenAI says this marks the first time an AI has cracked the top three in a major programming and math competition. OpenAI hasn’t revealed which specific model it used. The competition itself was sponsored by the company.