# Models

The foundation models you can run on and create endpoints on, how to choose a base, and how an upgrade moves an endpoint to a newer base.

> Informational documentation, based on the paper Data Language Models: A New Foundation Model Class for Tabular Data (arXiv:2605.06290, https://arxiv.org/abs/2605.06290). Reference material, not a hosted service; hosts, keys, and ids are placeholders.

Data and Endpoints are the objects you create; every self-serve operation is inference. "Model" always means a base model.

The model registry is the documentation of record for every foundation model, current and retired: intended use, training data, evaluation, lifecycle status, and known limitations.

## Choosing a base

Use the latest foundation model: it is what runs and new endpoints use when `base` is omitted, and it supports all three prediction task types (classification, regression, anomaly). Name a specific base only when you need a fixed comparison point.

A retired model answers runs, creations, and serves with `422 capability_unavailable`. An endpoint created on it keeps that base until you move it with `POST /v2/endpoints/:id/upgrade` (see [Endpoints](/model/endpoints#upgrade-a-new-base-shipped)); its earlier reports keep their base name, so the trail stays readable.

Bases are named `model-{n}` (`s{n}` in report names); the API version `/v2/` is a separate axis. See Identifiers.