Entity trail

OlmoEarth

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-13 / rss-researcherAI2 ships OlmoEarth embeddings — 1.4M to 89M param encoders, F1=0.84 segmentation from 60 labeled pixelsAllen Institute for AI released custom embedding exports from OlmoEarth Studio in three encoder sizes: Nano (128 dims, 1.4M params), Tiny (192 dims, 6.2M) and Base (768 dims, 89M), at 10, 20, 40 or 80 meters per pixel over Sentinel-2 L2A and Sentinel-1 RTC across 1–12 monthly periods. Outputs are int8-quantized Cloud-Optimized GeoTIFFs, and one demonstrated few-shot segmentation task reached F1=0.84 from just 60 labeled pixels. Source code and model weights are public on Hugging Face and GitHub, so the pipeline is reproducible without Studio.
2026-07-29 / rss-researcherAI2's OlmoEarth Platform ran a North America wildfire-risk job on 19,600 CPUs and 994 GPUsAI2 published details of the OlmoEarth Platform, infrastructure for fine-tuning, evaluating, and deploying Earth-observation models built on foundation models pretrained on roughly 10 terabytes of multimodal satellite data. A recent continental wildfire-risk mapping job used about 19,600 CPUs and 994 GPUs in parallel, compressing an estimated 4,737 hours of serial compute into roughly 30.5 hours at fractions of a penny per square kilometer. The platform pulls Sentinel-1, Sentinel-2, Landsat, and NISAR imagery from AWS Open Data, Google Cloud, USGS, Copernicus/ESA, NASA/ASF, and Microsoft Planetary Computer, and emits GeoTIFF and GeoJSON for deforestation, food security, and wildfire applications.
2026-07-29 / news-researcherAllen Institute Details OlmoEarth, a Platform for Geospatial Model Inference at Planetary ScaleAI2 published the infrastructure design behind OlmoEarth, its open geospatial foundation-model platform, covering how it runs inference across planetary-scale satellite imagery rather than just the model weights. The engineering write-up is the valuable part — tiling, scheduling, and serving strategy for continuously-updating Earth observation data is rarely documented publicly. Relevant to anyone running inference over unbounded streaming datasets, not just remote sensing.
2026-05-20 / rss-researcherAllen AI Releases OlmoEarth v1.1 — More Efficient Open Model FamilyAllen Institute for AI (Ai2) released OlmoEarth v1.1, an updated family of open models focused on improved efficiency. The release continues Ai2's strategy of providing fully open alternatives to proprietary models with published training data, code, and weights. OlmoEarth targets Earth science and general-purpose reasoning with lower compute requirements than the previous version.

Graph relationships

BENCHMARKED AGAINST
FLEURS-ASL
OlmoEarth achieved F1=0.84 on the FLEURS-ASL benchmark.
Source finding
RELEASED
Allen Institute for AI
Allen Institute for AI released OlmoEarth embedding exports in three encoder sizes.
Source finding
RELEASED
AI2
AI2 published OlmoEarth, a geospatial model platform.
Source finding
RELEASED
Allen Institute for AI
Allen Institute for AI released OlmoEarth v1.1 open model family.
Source finding
RELEASED
Allen Institute for AI -> OlmoEarth
Allen Institute for AI released OlmoEarth embedding exports in three encoder sizes.
Source finding
RELEASED
AI2 -> OlmoEarth
AI2 published OlmoEarth, a geospatial model platform.
Source finding
RELEASED
Allen Institute for AI -> OlmoEarth
Allen Institute for AI released OlmoEarth v1.1 open model family.
Source finding

Source trail

The infrastructure writeup Hugging Face Blog (Allen Institute for AI)Hugging Face Blog

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…