List your company
ResearchFormerly WizWand

Tabular Data Learning on TabArena-Lite

1,651Elo RatingTabICLv2-Retouche
1,315.081,402.291,489.51,576.71May 7, 2026
Updated 5mo ago

Evaluation results

MethodLinks
TabICLv2-RetoucheMode=T+E (Tuned + Ensembled), HPO budget=20x smaller (10 random configurations), Hardware=NVIDIA A10 / A100, Training time per 1K samples (s/K)=243.17, Prediction time per 1K samples (s/K)=22.032026.051,65166.510.2
TabICLv2-RetoucheMode=T (Tuned), HPO budget=20x smaller (10 random configurations), Hardware=NVIDIA A10 / A100, Training time per 1K samples (s/K)=243.17, Prediction time per 1K samples (s/K)=7.332026.051,63265.911
TabPFN-2.6Mode=D (Default), Hardware=NVIDIA H100, Training time per 1K samples (s/K)=5.75, Prediction time per 1K samples (s/K)=0.602026.051,62764.311.3
RealTabPFN-2.5Mode=T+E (Tuned + Ensembled), Hardware=NVIDIA H100, Training time per 1K samples (s/K)=2059.94, Prediction time per 1K samples (s/K)=9.792026.051,6146311.8
TabICLv2-RetoucheMode=D (Default), Hardware=NVIDIA A10 / A100, Training time per 1K samples (s/K)=20.80, Prediction time per 1K samples (s/K)=7.242026.051,60862.312.1
TabICLv2Mode=D (Default), Hardware=NVIDIA H100, Training time per 1K samples (s/K)=4.01, Prediction time per 1K samples (s/K)=0.352026.051,59563.312.7
RealMLPMode=T+E (Tuned + Ensembled), Training time per 1K samples (s/K)=2791.97, Prediction time per 1K samples (s/K)=13.892026.051,50045.817.7
LightGBMMode=T+E (Tuned + Ensembled), Training time per 1K samples (s/K)=416.56, Prediction time per 1K samples (s/K)=2.242026.051,40027.524
CatBoostMode=T+E (Tuned + Ensembled), Training time per 1K samples (s/K)=1665.53, Prediction time per 1K samples (s/K)=0.562026.051,38829.124.7
XGBoostMode=T+E (Tuned + Ensembled), Training time per 1K samples (s/K)=700.96, Prediction time per 1K samples (s/K)=1.442026.051,36023.126.7
TabPFNv2Mode=T+E (Tuned + Ensembled), Training time per 1K samples (s/K)=2942.08, Prediction time per 1K samples (s/K)=17.372026.051,32831.228.9