Hugging Face AI Updates: August 13, 2026

1. Ai2 Adds Custom Embedding Exports to OlmoEarth Studio

Hugging Face. The Allen Institute for AI published OlmoEarth embeddings on the Hugging Face blog, adding the ability to compute and export embedding vectors from Earth observation data in OlmoEarth Studio. Users can configure area of interest, time span across one to twelve monthly periods, encoder size (Nano, Tiny, or Base), spatial resolution from 10 to 80 meters, and Sentinel-2 and Sentinel-1 imagery sources, then download the results as Cloud-Optimized GeoTIFFs stored as signed 8-bit integers. The write-up demonstrates similarity search, few-shot land-cover segmentation reaching an F1 score of 0.84 from 60 labeled pixels, wildfire burn-scar change detection, and PCA-based exploration, with source code and weights on GitHub and Hugging Face. Source

2. Liquid AI Releases LFM2.5-VL-3B, a Compact Vision-Language Model for the Edge

Hugging Face. Liquid AI released LFM2.5-VL-3B, a 3.1-billion-parameter vision-language model for edge devices, on the Hugging Face Hub. The model pairs a SigLIP2 vision encoder with a text backbone and improves digital screen and UI comprehension, object detection and grounding, multi-image reasoning, and function calling over the prior version. Liquid AI reports 228 tokens per second on Apple’s M5 Max and 116 tokens per second on AMD’s Ryzen AI Max+ 395 while fitting in roughly 3GB of memory, and ships support for llama.cpp, MLX, vLLM, SGLang, and ONNX plus a browser-based WebGPU demo. Source