Python SDK Changelog

3.3.0 (2026-06-29)

get_model_run_output transparently waits for an output that is still being computed, polling until it is ready instead of failing fast.
Raises a clear error if the output is still computing after the poll timeout (default 300s).
Exposes Model.input_aggregations (list[InputAggregation] of id, name, key).

ModelSummary now includes a required revision field.
ModelGroup now exposes timeHorizon and a models list (id, revision, createdAt).
display_name values no longer include the .r<n> revision suffix; read revision for the revision number.
Model.summary exposes the terse model summary; description now carries long-form markdown. After the migration, description is an empty string for existing models until a long-form description is published; the previous terse text lives in summary.

Internal network requests use the new API gateway. This is not a true breaking change, but 3.0.0 will be the minimum version required in the future when the previous paths are deprecated.
The client now retries rate-limited responses, so bulk workloads self-throttle instead of failing on the first rate-limit response. A max_retries client option (default 3) tunes or disables the behaviour; set it to 0 to manage retries yourself.
A new RateLimitError (subclass of ApiError) is raised when retries are exhausted, with a retry_after attribute carrying the server’s hint so you can back off without guessing.

Multiple Client instances in the same process now share one auth token resulting in less authorisation calls.

Add has_aggregation_errors to ModelRun now so callers can detect runs whose input aggregations failed to process.

Model.notes changed to str | None (previously str) so you can branch when they’re absent. Note that this field is only populated for models authored by your organisation.
Add description: str to Model type — exposes the new top-level model description from the API.
Breaking: Remove the high specificity ModelMetadata Pydantic class. model.model_metadata is now dict[str, Any] and contains the arbitrary bag of metadata specific to the model. If you were previously reading this field, use model.model_metadata["<field>"], and remove anyfrom pd4castr_api_sdk.models import ModelMetadata imports.