TabPFN-3 advances tabular data prediction with speed and scale

A new technical report introduces TabPFN-3, an advanced foundation model for tabular data that significantly enhances performance and speed. This model scales to datasets with up to 1 million training rows and offers substantial reductions in training and inference times compared to its predecessor, TabPFN-2.5. TabPFN-3 demonstrates state-of-the-art results on various benchmarks, including TabArena, and shows improved capabilities in time series, relational, and tabular-text data prediction. AI

IMPACT Sets new SOTA on tabular benchmarks, potentially accelerating adoption in scientific and industrial prediction tasks.

RANK_REASON The cluster contains a technical report detailing a new model release and benchmark results for tabular data. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• AutoGluon 1.5
• fev-bench
• Léo Grinsztajn
• RelBenchV1
• TabArena
• TabPFN-2.5
• TabPFN-3
• TabPFN-3-Plus
• TabSTAR

• model release
• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →