Available on Baseten
AWS SageMaker
Open source
Code and weights.
Apache 2.0. Commercial use allowed.
Open-weight tabular foundation model
Nori V1.
Nori V1 is the open-weight foundation model for tables. Apache 2.0 code and weights, free for commercial use.
Model at a glance.
ExplainableBuilt-in interpretability
100M+ rowsScales to real-world data
No feature engineeringRaw tables in
No training or tuningNo hyperparameter search
Structured + unstructuredTables and text together
Quickstart
One call to predictions.
Pass test rows with labeled context. Nori returns predictions without fitting a new model.
pipinstallsynthefy-noriWeights download once and are cached locally.
from synthefy_nori import NoriRegressor model = NoriRegressor()predictions = model.predict(X_test, X_train, y_train)predict() takes test rows and labeled context in one call. Single GPU with automatic CPU fallback.Benchmarks
Beats tuned XGBoost / LightGBM zero-shot.
No feature engineering, training, or hyperparameter tuning.
Read the full evaluationFrontier performance at just 30M parameters
Nori-30M thinking beats Google’s 1.6B TabFM on median R².
Median R²0.8500.8550.8600.8650.8700.8750.880
Frontier performance at just 30M parameters
Nori-30M rivals Google’s 1.6B TabFM at roughly 50× smaller.
Combined Elo7008009001000110012001300
Run Nori
Open source or managed.
Managed API
No infrastructure to run.
Hosted through Baseten with SOC 2 Type II and zero data retention.
API quickstart