Releases·Model Release
Prior Labs releases TabPFN-3.5, SOTA tabular foundation model topping TabArena and BeyondArena
Prior Labs released TabPFN-3.5, a tabular foundation model that handles up to 1M rows and 20,000 features and tops the TabArena and BeyondArena benchmarks. It ships in three variants — Fast, Thinking, and Plus — trading speed against accuracy. The company is hosting a hackathon encouraging agents to use TabPFN-3.5 as a specialized prediction tool.
19 engagement3 sourcesTue, Sep 15, 2026, 04:18 PM
AI Overview
- Prior Labs released TabPFN-3.5, SOTA for predictions over 1M rows and 20k features with no domain-specific training, atop TabArena and BeyondArena (per a Prior Labs employee).
- TabPFN-3.5 leads BeyondArena on text-rich, high-cardinality data by +250 Elo over the strongest prior baseline.
- TabPFN-3.5-Fast (alpha) runs 6x faster than the base model; TabPFN-3.5-Thinking via API adds +20/+44 Elo.
- TabPFN-3.5-Thinking gains +20 Elo on BeyondArena and +44 Elo on TabArena over the base model by trading compute for accuracy.
- Prior Labs is hosting a hackathon for agent submissions using TabPFN-3.5 as a tabular-prediction tool, awarding NVIDIA GPUs to the top 3 entries.
Related
Releases·Model Release
NVIDIA releases Kumo Tabular, an open tabular foundation model ranking first on TabArena, BeyondArena, TALENT, and ScoringBench
NVIDIA released Kumo Tabular, an open foundation model for tabular classification and regression that predicts labels in a single forward pass with no training, tuning, or feature engineering. Weights and code are on Hugging Face and GitHub under the OpenMDW-1.1 license, and it ranks first on TabArena, BeyondArena, TALENT, and ScoringBench. It challenges the two-decade dominance of gradient-boosted trees in enterprise tabular ML.
NVIDIA released Kumo Tabular, an open foundation model for tabular classification and regression that predicts labels in a single forward pass with no training, tuning, or feature engineering. Weights and code are on Hugging Face and GitHub under the OpenMDW-1.1 license, and it ranks first on TabArena, BeyondArena, TALENT, and ScoringBench. It challenges the two-decade dominance of gradient-boosted trees in enterprise tabular ML.
1 engagement2 sourcesrss, redditTue, Sep 29, 2026, 03:30 PM
Developments·Paper
GIST: Generator-state-Informed Supervision Cuts Pretraining Variance in Tabular Foundation Models
An arxiv paper introduces GIST (Generator-state-Informed Supervision), a pretraining method for tabular foundation models built on prior-data fitted networks. Instead of sampled hard query labels, it averages query losses over conditional label distributions using intermediate generator states. The approach improves benchmark results without architecture or inference changes.
An arxiv paper introduces GIST (Generator-state-Informed Supervision), a pretraining method for tabular foundation models built on prior-data fitted networks. Instead of sampled hard query labels, it averages query losses over conditional label distributions using intermediate generator states. The approach improves benchmark results without architecture or inference changes.
0 engagement1 sourcearxivFri, Oct 9, 2026, 02:52 AM
Releases·Model Release
LG AI Research releases EXAONE Tabular 1.0, a tabular foundation model for in-context learning
LG AI Research released EXAONE Tabular 1.0, a compact tabular foundation model family for classification and regression via in-context learning, eliminating the need for dataset-specific gradient updates. The model is pretrained exclusively on a synthetic structural-causal-model (SCM) prior and introduces an architecture-centered redesign of tabular in-context learning.
LG AI Research released EXAONE Tabular 1.0, a compact tabular foundation model family for classification and regression via in-context learning, eliminating the need for dataset-specific gradient updates. The model is pretrained exclusively on a synthetic structural-causal-model (SCM) prior and introduces an architecture-centered redesign of tabular in-context learning.
0 engagement1 sourcearxivWed, Aug 26, 2026, 01:17 PM
Releases·Model Release
Google Research releases TabFM, a zero-shot foundation model for tabular data
Google Research introduced TabFM, a foundation model designed for zero-shot inference on tabular data. The model aims to handle diverse tabular tasks without task-specific fine-tuning, potentially simplifying machine learning workflows for structured data.
Google Research introduced TabFM, a foundation model designed for zero-shot inference on tabular data. The model aims to handle diverse tabular tasks without task-specific fine-tuning, potentially simplifying machine learning workflows for structured data.
64 engagement1 sourceredditFri, Jul 10, 2026, 06:42 PM
Developments·Paper
Mitra-v2 technical report: 77M-parameter synthetic-data tabular foundation model matches 1.6B TabFM
A technical report introduces Mitra-v2, a tabular foundation model claiming state-of-the-art performance on real-world classification and regression tasks including credit-risk scoring, clinical prediction, and house-price estimation. It is trained entirely on synthetic data using a small 2D Transformer backbone with longer contexts and larger feature spaces than Mitra-v1.
A technical report introduces Mitra-v2, a tabular foundation model claiming state-of-the-art performance on real-world classification and regression tasks including credit-risk scoring, clinical prediction, and house-price estimation. It is trained entirely on synthetic data using a small 2D Transformer backbone with longer contexts and larger feature spaces than Mitra-v1.
0 engagement1 sourcearxivThu, Sep 3, 2026, 10:55 PM