---
title: "Foundation Models"
description: "2 posts about Foundation Models, summarised, each linking to the original."
---

# Foundation Models
> 2 posts about Foundation Models, summarised, each linking to the original.

## Articles

### [OlmoEarth v1.3: State-of-the-art embeddings for the whole planet](https://yomu.fyi/post/olmoearth-v1-3-state-of-the-art-embeddings-for-the-whole-planet.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Kyle Wiggers
- Published: Oct 22, 2026
- Type: Announcement

OlmoEarth v1.3 is a satellite embedding model that reads a year of radar and multispectral time series and outputs a 128-dimensional vector per 10-meter pixel, with the first 64 dimensions usable as a compact variant. It scores highest on 20 of 27 evaluations across six continents, outperforming AEF and Tessera v2, and generalizes best on the new PLANTEUR crop-type benchmark. Precomputed global embeddings at 10-meter resolution for 2017–2025 covering land and coastal waters, along with model weights, data, and code, are fully open on Source Cooperative, enabling anyone to track change or compare places without a GPU. Training used 53% of v1.2’s compute and 18% of v1’s, with 50% and 22% of their carbon emissions, while matching v1.2’s fine-tuning performance. The OlmoEarth Embeddings Explorer provides a browser-based false-color map to compare years, find similar locations, and train classifiers on the spot without an account.


### [NVIDIA Kumo Relational Outperforms Supervised Machine Learning on Relational Data, No Training Required](https://yomu.fyi/post/nvidia-kumo-relational-outperforms-supervised-machine-learning-on-rela.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Joshua Przybylko, Jure Leskovec, Matthias Fey, Federico Lopez, Valter Hudovernik, Akihiro Nitta, Ramona Bendias, Ardrian Wong
- Published: Oct 21, 2026
- Type: Announcement

NVIDIA Kumo Relational is a pre-trained relational foundation model that makes predictions directly on enterprise databases without requiring task-specific training, feature engineering, or table flattening. Given a predictive query, it gathers up to 10,000 historical examples as context and produces outcomes such as churn or demand in a single forward pass. Across 41 tasks on 15 databases, it outperforms supervised relational deep learning models and gradient-boosted tree pipelines; on the SAP SALT benchmark it surpasses AutoGluon ensembles by 8% in mean reciprocal rank. The model scales to databases with over 500 billion rows and is available through NVIDIA’s open-source structured-data-models library or as an NVIDIA NIM. It reasons within tables using column and row attention and across tables using graph attention over foreign keys, avoiding quadratic cost.