OpenAI and Google’s AI models achieved impressive results in a difficult math competition, but disputed how the other got their score.