ORC wrote these orders 12 percent smaller than Parquet 129 and keeps richer statistics (sums, per-10,000-row indexes), but Parquet won the ecosystem.
| Aspect | Parquet | ORC |
|---|---|---|
| Native home | Spark 129 , lakehouse tables, DuckDB 61,228 , Python | Hive 129 , Trino 403,499 and Spark on Hive tables |
| Nested data | Repetition and definition levels | Separate child columns with lengths |
| Statistics | Per row group, optional page index | Per file, stripe and 10,000-row group |
| Iceberg 129 and Delta support | Default (Delta: Parquet only) | Supported by Iceberg |
Pick ORC when the platform is Hive-centric (Hive's ACID tables require it); otherwise choose Parquet, which every engine, library and cloud warehouse reads natively.