Download the scores, or query them with the Scores API

Mon, 02 Nov 2026 00:00:00 GMT

The scores displayed on the Scores page can now be downloaded, for every published OceanBench version, region and track.

Export the displayed scores

An Export row sits next to the Region and Track pickers.

The Export row of the Scores page.

Each button opens a popup for the selected version, region and track, with:

a button to download the file, or to copy the Zarr store URL,
code snippets to open the scores in Python or download them from a shell,
for CSV and Parquet, a link to explore the scores in the EDITO Data Explorer, without installing anything.

The Zarr popup and its Python snippet.

Three formats

CSV and Parquet hold one row per challenger, comparison, metric, depth, variable and lead day, ready for pandas or a spreadsheet.
Zarr holds a multi-dimensional dataset: value over the challenger, comparison, metric, depth, variable and lead_day dimensions, with the unit and standard_name of each variable. Select any slice with xarray:

import xarray

scores = xarray.open_zarr(
    "https://oceanbench.edito.eu/api/scores//global/high_resolution.zarr",
    consolidated=True,
)
scores["value"].sel(challenger="glonet", comparison="reanalysis", depth="Surface")

The Scores API

The same files are served at stable URLs, so scripts and pipelines can fetch them directly:

curl https://oceanbench.edito.eu/api/scores/index.json
curl -O https://oceanbench.edito.eu/api/scores/<version>/<region>/<track>.parquet

index.json lists every available version, region and track. The Scores API documentation describes the endpoints, columns and dimensions.