explainx / blog
Google Research''s TabFM (June 30, 2026) brings in-context learning to tabular ML — one forward pass on unseen tables, no hyperparameter tuning or feature engineering. Hybrid TabPFN + TabICL architecture, hundreds of millions of synthetic training sets, TabArena Elo leader, BigQuery AI.PREDICT coming soon.
Jun 19, 2026
TimesFM 2.5 packs a 16k context window and continuous quantile forecasting into 200M parameters. Available on PyPI, Hugging Face, BigQuery ML, and Google Sheets. Here's how it works.
Jun 22, 2026
The Swiss AI Initiative just released Apertus — a fully open 8B and 70B foundation model with open weights, open data, and full training recipes. Built by EPFL, ETH Zurich, and CSCS, it's the most transparent large language model attempt to date and a direct answer to AI sovereignty concerns in Europe and beyond.
Jul 2, 2026
diegosouzapw/OmniRoute hit ~9.8k GitHub stars with v3.8.42 — route every coding agent through localhost:20128/v1, auto-fallback across subscriptions and free providers, and stack compression to stretch quotas. MIT, npm, Docker, desktop.
Tabular data still runs most of enterprise ML — churn prediction, fraud detection, credit scoring, ops forecasting on structured columns. For decades the workflow was the same: load a table, engineer features, tune XGBoost or a random forest, cross-validate, deploy.
Google Research's TabFM (announced June 30, 2026) asks a different question: what if tabular prediction worked like a language model — zero-shot, one forward pass, no per-dataset training?
It is the tabular sibling to TimesFM, which already shifted how teams handle time-series forecasting. TabFM targets classification and regression on mixed-type columns with a scikit-learn-compatible API, weights on Hugging Face, code on GitHub, and pip install tabfm (v1.0.0).
| What it is | Foundation model for tabular classification + regression via in-context learning |
| Release | June 30, 2026 — TabFM v1.0.0 |
| Paradigm | Entire dataset (train rows + test rows) as one prompt — no weight updates per task |
| Architecture | Hybrid of TabPFN-style row/column attention + TabICL-style compressed row ICL |
| Training | Hundreds of millions of synthetic tables from structural causal models |
| Benchmark | Top TabArena Elo vs tuned tree ensembles and tabular DL baselines |
| Variants | TabFM (single pass) · TabFM-Ensemble (32-way + NNLS weights + Platt scaling) |
| Deploy paths | PyPI · Hugging Face · BigQuery AI.PREDICT (coming weeks) |
| License | Apache 2.0 (not an officially supported Google product) |
Tree models — AdaBoost, XGBoost, random forests — still dominate structured data benchmarks because they handle heterogeneous columns, missing values, and nonlinear interactions well. But deployment cost is high:
LLMs showed a different pattern: in-context learning (ICL) — show examples in the prompt, get answers without fine-tuning weights. TabFM applies that idea to two-dimensional, orderless tables where swapping rows or columns should not change meaning.
Traditional supervised ML updates parameters to match each dataset's distribution. TabFM does not. At inference time it receives:
…as a unified context. The model learns column relationships and row patterns from that context in one pass.
Tables are not natural language — you cannot naively tokenize a CSV like a sentence. TabFM's architecture (from the official blog) combines ideas from TabPFN and TabICL:
Raw table values pass through multilayer attention that alternates across columns (features) and rows (examples) — similar to TabPFN. This captures feature interactions without hand-built crosses — the work data scientists usually do manually.
After cross-attention, each row's representation compresses into a single dense vector — reducing the 2D grid to a 1D sequence of row embeddings.
A Transformer attends over compressed row vectors (TabICL-style), not the full uncompressed grid. That keeps compute manageable on larger tables while preserving zero-shot ICL behavior.
Order invariance: The design respects that permuting rows or columns should not change the underlying prediction task — unlike sequential text.
Real industrial tabular datasets are proprietary, schema-locked, and scarce at foundation-model scale. TabFM's pretraining recipe:
This mirrors how other tabular foundation models (TabPFN, etc.) lean on synthetic pretraining — but at Google's stated massive scale.
Google evaluated on TabArena — a living benchmark system using Elo ratings from head-to-head win rates:
| Scope | Coverage |
|---|---|
| Classification | 38 datasets |
| Regression | 13 datasets |
| Sample sizes | 700 → 150,000 rows |
Two configurations shipped:
| Variant | What it does | Tuning required |
|---|---|---|
| TabFM | Single forward pass, out-of-the-box | None |
| TabFM-Ensemble | Cross features + SVD features, 32-way ensemble, NNLS optimal weights; Platt scaling on classification | Ensemble setup, not full HPO on base trees |
On TabArena Elo plots, both variants sit at or above heavily tuned industry baselines — including default and tuned+ensemble tree pipelines (labeled (D) and (T+E) on Google's charts).
Per-fold metrics and head-to-head win rates vs specific baselines live on the GitHub repo.
TabFM v1.0.0 is scikit-learn compatible. Weights auto-download from Hugging Face:
pip install tabfm
from tabfm import tabfm_v1_0_0
# Load pretrained TabFM v1.0.0 (JAX or PyTorch backend)
model = tabfm_v1_0_0.load()
# Standard sklearn-style API — ICL happens inside predict
model.fit(X_train, y_train)
predictions = model.predict(X_test)
Requirements (from repo):
jax==0.10.1, flax==0.12.7) or PyTorch (torch==2.12.1+) backendSee google-research/tabfm for ensemble configs, TabArena reproduction, and pinned requirements.txt.
AI.PREDICT Coming SoonGoogle is integrating TabFM into BigQuery — in the coming weeks, teams will run advanced regression and classification via:
-- Conceptual — exact syntax when GA ships
SELECT AI.PREDICT(MODEL tabfm, ...) AS prediction
FROM your_table;
That mirrors the TimesFM → BigQuery ML path: SQL-native access for analysts who never touch Jupyter. No ML expertise required for basic predictive workflows on warehouse data.
| Approach | Per-dataset training | Feature engineering | Typical strength |
|---|---|---|---|
| XGBoost / RF | Yes — full fit + HPO | Often extensive | Strong default on medium tables |
| TabPFN | No — ICL | Minimal | Small/medium tables, fast ICL |
| TabICL | No — compressed ICL | Minimal | Efficient ICL at scale |
| TabFM | No — hybrid architecture + massive synthetic pretrain | Eliminated for zero-shot | TabArena-leading Elo, sklearn API, BigQuery path |
| LLM on CSV text | Prompt-only | Fragile on wide/numeric tables | General reasoning, not tabular-native |
TabFM's pitch is foundation-model convenience — like grabbing TimesFM for a new time series without retraining — applied to the most common enterprise ML substrate.
Data scientists tired of HPO loops — TabFM is a credible zero-shot first pass before investing in XGBoost tuning.
ML platform teams — sklearn API + Hugging Face weights + impending BigQuery SQL lowers integration friction.
TimesFM users — If forecasting already moved to foundation models, tabular is the natural next dataset type in the same Google Research stack.
Enterprise analytics — Warehouse-native AI.PREDICT could put tabular FM inference next to SQL dashboards without a separate training pipeline.
pip install tabfm