Open LimiX Studio

General-purpose data foundation models
for trustworthy decision-making.

Explore

Stable AI / LimiX-2

Foundation modelfor structured data

LimiX-2 is our flagship foundation model for classification, regression, and missing-value imputation, enabling direct inference without task-specific training.

Try LimiX Studio
Explore LimiX-2
Illustrative example: context rows help predict a missing target.

Benchmark evidence / updated 18 Sep 2026

LimiX-2 achieves leading performance on major tabular benchmarks

Top five methods · point estimates + published 95% intervalsOverall Elo rating ↑

  1. LimiX-2Default · Foundation Model1943−80 / +118
  2. TabPFN-3.5Default · Foundation Model1861−64 / +94
  3. TabFM+Open-source system1821−83 / +108
  4. CausiloDefault · Foundation Model1785−55 / +90
  5. AutoGluon 1.6Noncommercial · 4h · Open-source system1784−58 / +98

Official TabArena models + open-source systems pool (entrants=open), updated 18 Sep 2026 and accessed 20 Sep 2026.

TabArena · Models + open-source systems · 18 Sep 2026
Evaluation methodology Current live TabArena

Core capabilities

One model for many tasks

Released task

Classify new observations

Use labeled examples in the table as context to infer the class of each new row—without fitting a dataset-specific model.

Context
Labeled feature–target examples
Output
Class probabilities for each query row
Observed context LimiX-2
RegionTemp.FlowPowerMode
01North18.61.242.8Idle
02East24.12.473.2Active
03West21.51.858.4Active
QuerySouth23.72.167.5?Active
Inference output
Predicted classActive
In-context inferenceOne pretrained checkpoint

Application areas

From Data to Decision

Application areas

Make production data actionable

  • Equipment fault prediction
  • Process optimization
  • Quality prediction
01Sensor table
02LimiX
03Failure risk / quality score
Abstract workflow · examples are application areas, not deployment claims.
01Manufacturing

Application areas

Make production data actionable

  • Equipment fault prediction
  • Process optimization
  • Quality prediction
01Sensor table
02LimiX
03Failure risk / quality score
Abstract workflow · examples are application areas, not deployment claims.
02Energy

Application areas

Turn operating data into clearer system signals

  • Energy forecasting
  • Equipment maintenance
  • Dispatch optimization
01Grid records
02LimiX
03Forecast / operating signal
Abstract workflow · examples are application areas, not deployment claims.
03Telecommunications

Application areas

Read patterns across complex network operations

  • Network fault prediction
  • Congestion analysis
  • User value modeling
01Network table
02LimiX
03Risk / demand estimate
Abstract workflow · examples are application areas, not deployment claims.
04Finance

Application areas

Support decisions across incomplete, changing records

  • Risk classification
  • Fraud screening
  • Demand estimation
01Financial records
02LimiX
03Risk / continuous estimate
Abstract workflow · examples are application areas, not deployment claims.
05Healthcare

Application areas

Learn from heterogeneous clinical observations

  • Condition classification
  • Outcome estimation
  • Missing-observation recovery
01Clinical table
02LimiX
03Class / estimate / recovery
Abstract workflow · examples are application areas, not deployment claims.

Research and collaboration

Research you can inspect and build on

01

Research

Questions before claims

  • Structured-data scaling
  • Mechanism-oriented modeling
  • Causal structure
Explore Research
02

Open models

Inspect the work

  • LimiX-2
  • GitHub
  • Hugging Face
  • Technical Report
Explore LimiX
03

Industry collaboration

Bring research into practice

  • Energy
  • Telecommunications
  • Industrial systems
  • Publicly documented projects
Work with Stable AI
LimiX-2 Technical Report

From one tableto better decisions

Explore LimiX-2Developer GuideWork with Stable AI