TabPFN-3.5 Outperforms Tuned GBDTs in Telecom Churn Profitability
November 11, 2026
TabPFN-3.5 earns higher expected maximum profit than Optuna-tuned GBDTs on telecom churn datasets. Specifically, it shows profit gains of 1.9% on Cell2Cell and 7.6% on Iranian datasets compared to the best tuned tree ensembles.
HOW THIS AFFECTS YOU
●
builderYou can use TabPFN-3.5 for tabular prediction tasks where profit or calibration is more important than simple ranking.
●
founderUsing these foundation models for churn might yield better financial returns than traditional tuned gradient boosting.