AI Potluck
Back to Gap Map Model components / Scientific AI models

Prithvi WxC

NASA IMPACT
open weights / Overall score: 2.4

Prithvi WxC is a NASA-IBM weather and climate foundation model trained on 160 MERRA-2 variables. Its 2.3-billion-parameter architecture combines masked reconstruction with forecasting, accepts flexible time intervals, and has separate base and rollout checkpoints plus downstream adaptations for regional forecasting and climate processes.

Verified 2026-09-10 via the Hugging Face model card and NASA IMPACT repository.

Openness

3 high confidence
3.0
license
CDLA-Permissive-2.0(permissive non-OSI)
weights
open(CDLA-Permissive-2.0 ungated checkpoint)
data
documented-not-released(MERRA-2 variables and climatology identified
code
partial(model, validation, and downstream task code without full pretraining pipeline)

The downloadable checkpoint has a permissive data-oriented license and no use ceiling; the missing pretraining pipeline keeps it at the open-weights center.

Adoption

1 high confidence
1.0

2,138 downloads in the trailing 30 days for the declared checkpoint.

Capability

3 high confidence
3.0

Two attributes. Generative is declined under ruling (d) - producing a future timestamp from two input timestamps is a deterministic forecast, which is prediction rather than generation, and the same reading declines aifs, pangu-weather, climax and aurora. 160 variables are not counted as multiple instruments.

Verified 2026-09-10