IBM-NASA Prithvi-EO
Prithvi-EO-2.0 is an open geospatial foundation model from IBM and NASA, released on Hugging Face on 4 December 2024 in 300M and 600M sizes. It is trained on Harmonized Landsat and Sentinel-2 data and fine-tunable through IBM TerraTorch, with no platform lock-in.
IBM-NASA Prithvi-EOInside IBM-NASA Prithvi-EO
Prithvi-EO-2.0 is a multi-temporal geospatial foundation model pretrained on roughly 4.2 million global HLS time-series samples at 30m, with temporal and location embeddings.
The 600M variant reported an 8 percent improvement over the prior Prithvi model and outperformed six other geospatial foundation models across benchmark tasks spanning 0.1m to 15m resolution.
It ships with open weights on Hugging Face and is fine-tunable through IBM TerraTorch, so it can be self-hosted and adapted to specific MRV tasks without depending on a vendor API.
An open, self-hostable foundation model for biomass, land-cover and change tasks, fine-tunable without depending on a vendor API.
Recent developments
- Dec 2024Prithvi-EO-2.0 (300M and 600M) released on Hugging Face, fine-tunable via TerraTorch.
Using IBM-NASA Prithvi-EO in a project
The provider’s own site holds its current capabilities, access terms and documentation. Read them before selecting the tool or planning an integration.