The following table shows the result of a small benchmark comparing several Python ORM for importing the 10000+ terms of the ICD10 medical classifications (writing), and then for traversing all terms (reading).

All ORM used SQLite3. In addition, plain SQL is also shown for comparison.

Results:

Module	Writing	Reading	Database size	Lines of code	Inheritance
SQLAlchemy 1.1.13	20.32 seconds	6.41 seconds	962 560 bytes	42 lines	simple (?)
SQLObjet 3.4.0	7.58 seconds	2.19 seconds	966 656 bytes	35 lines	simple
Peewee 2.10.1	5.37 seconds	2.08 seconds	962 560 bytes	35 lines	simple (?)
Pony 0.7.2	7.10 seconds	0.61 seconds	958 464 bytes	35 lines	multiple (partiel)
Owlready 2 0.3	1.41 seconds	0.85 seconds	4 235 264 bytes	38 lines	multiple (full)
ORMithorynque	1.65 seconds	0.49 seconds	1 269 760 bytes	31 lines	multiple (full)
SQL (non-objet)	1.12 seconds	0.19 seconds	966 656 bytes	45 lines	none

(for numbers, the lower is always the better)

ORMithorynque beats all other ORM for speed. It also require fewer lines of code (mostly thanks to the automatic schema creation and update).

ORMithorynque databases are slighly bigger than with other ORM however; this result was expected because ORMithorynque requires an extra table for storing classname, due to multiple inheritance support.