Prithvi WxC
2.3B-parameter open weather-and-climate foundation model trained on 40 years of MERRA-2 data.
About
Prithvi WxC is a weather-and-climate foundation model built jointly by IBM and NASA with contributions from Oak Ridge National Laboratory, announced in September 2024. The 2.3-billion-parameter model is pretrained on 40 years of NASA's MERRA-2 reanalysis across 160 atmospheric variables, using both forecasting and masked-reconstruction objectives — so it can propagate an atmospheric state forward in time or fill one in from partial observations. Its architecture is designed to fine-tune down to global, regional and local scales.
Everything is open: Apache 2.0 weights on Hugging Face (ibm-nasa-geospatial/Prithvi-WxC-1.0-2300M), code on GitHub, and a technical paper. IBM has also shipped fine-tuned versions for climate downscaling and gravity-wave parameterization. In the paper's experiments the model reconstructed global surface temperatures from just 5% of the original data, which hints at uses well beyond forecasting — data assimilation, downscaling global climate runs, and targeted local forecasts.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Paper-reported ("Prithvi WxC: Foundation Model for Weather and Climate"): accurate reconstruction of global surface temperatures from a random 5% sample of data
- No standardized public weather-benchmark leaderboard scores published
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (TIER metrics, metrics-v1). Evidence: Apache 2.0 weights + code on Hugging Face/GitHub (ibm-nasa-geospatial); joint IBM/NASA technical paper with downscaling and gravity-wave fine-tunes; institutional backing from two agencies plus ORNL. Strengths: fully open license, serious institutional maintenance, broad fine-tuning story. Docked for: (1) headline results are author-reported with no independent benchmark replication found; (2) the 28GB checkpoint and MERRA-2 data pipeline make self-hosting heavyweight. License: Apache 2.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · 2.3B params, 160 MERRA-2 variables, 40 years of data; announced Sep 2024
- · Apache 2.0 weights on HF (ibm-nasa-geospatial/Prithvi-WxC-1.0-2300M); code on GitHub
- · Fine-tunes for downscaling and gravity-wave parameterization published