> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/sdgm/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/sdgm/_mcp/server.

# Which foundation model should I choose?

> Choose a foundation model based on your data and prediction task

Choose by your data structure and prediction task:

* **Kumo Tabular:** Classification or regression on a single table of numerical and categorical features. Text, images, and timestamps need preprocessing into supported features; related tables and relationships are not direct inputs.
* **Kumo Relational:** Classification or regression using information across related tables. Check the model version and runtime documentation for supported tasks and outputs.
* **Kumo Forecast or Kumo Anomaly:** Time-series forecasting or anomaly detection, respectively.

Tabular and Relational need labeled examples as context even though they do not require task-specific training. Prediction quality depends on how well that context represents the rows you want to score. For Tabular, accuracy can also degrade on tables far outside the pretraining ranges.

See [Choose a Model](/about-nvidia-kumo/overview#choose-a-model) for model cards, required inputs, and execution options. Before deployment, [evaluate the model on your own data](/faq/kumo-foundation-models/how-do-i-evaluate-foundation-models). For memory requirements, see [Context-Size Limits](/faq/kumo-foundation-models/what-are-the-context-size-limits).