From an infrastructure perspective, this is a logical step. Supporting legacy generative models burns valuable GPU quotas in data centers. The Sora 2 architecture works more efficiently with tensor cores, consumes less power during rendering, and generates fewer artifacts. Users are given a window to export their old projects. OpenAI demonstrates a strict approach to its ecosystem: the company will not squander resources on backward compatibility if it hinders the implementation of new, more efficient pipelines.
Source: OpenAI Help Center
SoftwareOpenAISoraGenerative VideoUpdate