I haven't written benchmarks yet but it is something I am working on. The reason that performance is mentioned in the README is that I haven't spent time on improving it yet.

The biggest blockers in terms of performance are:

- Communication with the internal Rust engine is performed over HTTP instead of FFI.

- You cannot select a subset of fields (every scalar field in a model is selected)

- You cannot skip pydantic validation which while fast is unnecessary in some cases

With that said Prisma Python should be reasonably fast but I would expect it to be on the lower end of the spectrum compared to other Python ORMs. I have ran some load tests locally and creating 500,000 records took about 90 seconds.