Tabular Foundation Models
In-context learning on tables.
Tier Grades
Tier A: Frontier AI labs
feasible
high confidence
Pretraining needs about 8 GPUs for 2 weeks, trivial at frontier scale.
Tier B: Indonesian AI labs
feasible
high confidence
8 GPUs for 2 weeks on synthetic data is modest.
Tier C: Institution units such as ITB
feasible
high confidence
An academic group (ELLIS Tübingen) produced TabPFN on 8 GPUs, matching the academic 1 to 8 GPU range.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
high confidence
TabPFN is fine-tunable and answers in 2.8 s on up to 10,000 samples, suiting proof-of-value work on client tables.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeTabPFN: pretrained about 2 weeks on 8 GPUs; inference 2.8 s vs a 4-hour tuned ensemble.
- DataAbout 130M synthetic datasets from structural causal models; handles up to 10,000 samples and 500 features.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): not sourced in the notes.
- Time horizonMonths.
- Shelf-life signalsnot in the notes
- Pipeline roleTier C and academic groups produce tabular models; Tier B and Tier D apply and fine-tune them on sector tables.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
ELLIS TübingenAcademic groupLinked models
TabPFNtabularexemplarTabPFN-3.5tabularproduced by
TabFM 1.0tabularproduced by
Noritabularproduced byTabPFN-3tabularproduced byTabPFN-2.6tabularproduced by
TabICL v2tabularproduced byRealTabPFN-2.5tabularproduced by
SAP-RPT-OSS (sap-rpt-1-oss)tabularproduced by
TabSTARtabularproduced by
Mitratabularproduced byTabICLtabularproduced by
Orion-Bixtabularproduced byOrion-MSPtabularproduced byTabPFN v2tabularproduced byContextTabtabularproduced byLimiXtabularproduced by
TabDPTtabularproduced byTabPFN v1tabularproduced by
KumoRFMgraphproduced byTabPFN-TStime seriesused byLinks
R&D TopicTabular Foundation Models
ModelsTabPFNPrior Labs, tabularMitraAmazon, tabularTabICLINRIA, tabularTabICL v2INRIA, tabular+17 more
TabPFNPrior Labs, tabularMitraAmazon, tabularTabICLINRIA, tabularTabICL v2INRIA, tabular+17 moreUse Casesnone recorded
Who Runs Itnone recorded
Models 21
TabPFNPrior Labs, ELLIS Tübingen, tabularexemplarTabPFN v2Prior Labs, tabularproduced byTabPFN-2.6Prior Labs, tabularproduced byTabPFN-3Prior Labs, tabularproduced byTabPFN v1Prior Labs, tabularproduced byTabPFN-3.5Prior Labs, tabularproduced byKumoRFMKumo AI, graphproduced byRealTabPFN-2.5Prior Labs, tabularproduced byTabICLINRIA, tabularproduced byTabICL v2INRIA, tabularproduced byTabPFN-TSPrior Labs, time seriesused byContextTabSAP, tabularproduced by+9
LimiXStableAI, tabularproduced byMitraAmazon, tabularproduced byNoriSynthefy, tabularproduced byOrion-BixLexsi Labs, tabularproduced byOrion-MSPLexsi Labs, tabularproduced bySAP-RPT-OSS (sap-rpt-1-oss)SAP, tabularproduced byTabDPTLayer 6 AI, tabularproduced byTabFM 1.0Google, tabularproduced byTabSTARTechnion - Israel Institute of Technology, tabularproduced by
MitraAmazon, tabularproduced byNoriSynthefy, tabularproduced byOrion-BixLexsi Labs, tabularproduced byOrion-MSPLexsi Labs, tabularproduced bySAP-RPT-OSS (sap-rpt-1-oss)SAP, tabularproduced byTabDPTLayer 6 AI, tabularproduced byTabFM 1.0Google, tabularproduced byTabSTARTechnion - Israel Institute of Technology, tabularproduced byPapers
- TabPFN: A Transformer That Solves Small Tabular Classification Problems in a SecondNoah Hollmann et al., arXiv preprint, 2022 founding
- TabICL: A Tabular Foundation Model for In-Context Learning on Large DataJingang Qu et al., ICML 2025 (per arXiv comment), 2025 state of the art
- TabArena: A Living Benchmark for Machine Learning on Tabular DataNick Erickson et al., NeurIPS 2025 (per arXiv comment), 2025 state of the art
- Why do tree-based models still outperform deep learning on tabular data?Léo Grinsztajn et al., arXiv preprint, 2022 state of the art
- Tabular Data: Deep Learning is Not All You NeedRavid Shwartz-Ziv et al., arXiv preprint, 2021 survey
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- $100K or 100 Days:Trade-offs when Pre-Training with Academic Resources arxiv.org https://arxiv.org/html/2410.23261v1
- Client Challenge nature.com https://www.nature.com/articles/s41586-024-08328-6
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Tabular Foundation Models", https://aurion.semai.id/topics/tabular-foundation-models/, records of October 2026.