TabPFN-3.5 pushes foundation models deeper into tabular data
TabPFN-3.5 is now the default model in Prior Labs’ open-source package, with a smaller TabPFN-3.5-Fast variant for lower-latency inference. The repo supports local classification and regression through a few lines of Python, while the release targets much larger tables than earlier versions, up to 1 million rows and 20,000 features. That matters because tabular teams can test foundation-model inference without immediately rebuilding their stack around a hosted API.