Nums AI — Foundation models for structured data

We build foundation models for structured data. Predict what matters, directly from your data.

Open Weight

Cloud Deploy

API

Accuracy and speed, all in the same model

Our foundation model for structured data. Strong predictions, without task-specific training.

The two plots use the same model order. Accuracy bars show final TabArena Elo on a zero-based 0–2,000 axis. Time bars show recorded fit plus predict seconds per 1,000 total rows on a logarithmic 0.1–1,000 second axis. Shorter time bars mean less recorded compute cost; equal intervals on that axis mean ten times the cost. The values and positions stay final during the brief entry reveal; this is not a live benchmark or serving latency comparison.

Accuracy

Time to Prediction

Start with Python

Send a context table with the target column and new query rows. We host the model.

Quickstart

APIPython

# pip install causilo-client
import os
from causilo_client import Causilo

client = Causilo(
    "https://api.nums.world", token=os.environ["CAUSILO_KEY"]
)
predictions = client.predict(
    context, query, target="target", model="causilo-reg"
)

Run Causilo on your own infrastructure with a familiar scikit-learn interface.

Model Weights Source Code

Research and evaluation. Commercial use requires a separate license. Model License

Open WeightPython

# pip install causilo
from causilo import CausiloRegressor

model = CausiloRegressor()
model.fit(X_train, y_train)
predictions = model.predict(X_test)