Prithvi-EO
NASA IMPACTPrithvi-EO is an Earth-observation foundation-model family jointly developed by NASA and IBM. Its spatiotemporal vision transformer consumes Harmonized Landsat and Sentinel imagery together with location and acquisition-time metadata, and the release includes several model sizes plus configurations and notebooks for downstream remote-sensing tasks.
Openness
3 high confidence- license
- Apache-2.0(OSI)
- weights
- open(Apache-2.0 ungated checkpoint)
- data
- documented-not-released(4.2M HLS V2 samples and six bands described
- code
- partial(inference and downstream fine-tuning code, not the pretraining pipeline)
The weights are permissively licensed and the training inputs are described, but the repository starts at inference and downstream fine-tuning rather than reproducing pretraining.
- https://huggingface.co/ibm-nasa-geospatial/Prithvi-EO-2.0-600M/raw/main/README.md recorded 2026-10-13
The card identifies Apache-2.0 weights, 4.2M HLS V2 training samples with six bands, six released model variants, and TerraTorch fine-tuning examples.
Adoption
1 high confidenceAdoption is measured as trailing-30-day Hugging Face downloads of the declared 600M checkpoint.
- https://huggingface.co/api/models/ibm-nasa-geospatial/Prithvi-EO-2.0-600M recorded 2026-09-10
downloads: 1899; gated: false; private: false.
Capability
4 high confidenceThree of the four capability attributes are established, one below TerraMind; masked pretraining reconstruction does not by itself count as a scientifically meaningful generative output.
- https://raw.githubusercontent.com/NASA-IMPACT/Prithvi-EO-2.0/main/README.md recorded 2026-09-10
Documents HLS satellite time series, TerraTorch configurations and notebooks for downstream tasks, and evaluation from 0.1 m to 15 m resolution.
Verified 2026-09-10