snowflakedb/snowpark-python v1.56.0

Release

on GitHub

1.56.0 (2026-10-29)

Snowpark Python API Updates

New Features

• Added DataFrameReader.documents() (private preview) to read staged documents into DataFrames. Provide a schema to extract structured fields, or omit it to return parsed document content. Schema-based reads also return an EXTRACTION_SCORES column with model-reported scores from 0 to 1 for each top-level field. The reader can process a single file or a document directory, automatically configures token limits, and splits long documents across model calls. It also parallelizes processing across documents, and isolates single-file failures so other documents in the directory can continue processing.

Improvements

• Eager schema validation (context.configure_development_features(enable_eager_schema_validation=True)) now raises invalid column references in DataFrame.filter and DataFrame.sort when the DataFrame is created.

Bug Fixes

• Fixed a bug that the dataframe trace and missing-object information were missing from errors raised while a DataFrame is being created, when debuggability features are enabled by context.configure_development_features.
• Fixed Session.client_telemetry failing to obtain a workload-identity auth token on the v5 / universal-driver connector, which does not expose auth_class, _session_manager, or top-level wif_util.