# Parquet.mojo available for your columnar data needs

**URL:** <https://forum.modular.com/t/parquet-mojo-available-for-your-columnar-data-needs/3470>\
**Category:** Community Showcase\
**Created:** [September 6, 2026, 2:11pm UTC](https://forum.modular.com/t/parquet-mojo-available-for-your-columnar-data-needs/3470 "2026-09-06T14:11:40Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![mseritan](https://sea1.discourse-cdn.com/flex001/user_avatar/forum.modular.com/mseritan/32/130_2.png) [@mseritan](https://forum.modular.com/u/mseritan)\
**Post date:** [September 6, 2026, 2:11pm UTC](https://forum.modular.com/t/parquet-mojo-available-for-your-columnar-data-needs/3470/1 "2026-09-06T14:11:40Z")

</div>

I believe the [performance](https://magmalake.org/blog/parquet-mojo-against-pyarrow/) and [correctness](https://magmalake.org/blog/how-far-parquet-mojo-is-validated/) bars are high enough for parquet.mojo to be usable.

My main use case is to write whole processing pipelines in Mojo. If people want to use this from python I can add a wheel. Let me [know](https://github.com/magmalake/parquet.mojo/issues).

> **[GitHub - magmalake/parquet.mojo: Native Apache Parquet reader and writer for Mojo —...](https://github.com/magmalake/parquet.mojo)**
>
> Native Apache Parquet reader and writer for Mojo — all encodings and codecs, Arrow C Data Interface output

---

<div class="post-metadata">

**Author:** ![mseritan](https://sea1.discourse-cdn.com/flex001/user_avatar/forum.modular.com/mseritan/32/130_2.png) [@mseritan](https://forum.modular.com/u/mseritan)\
**Post date:** [September 8, 2026, 2:22am UTC](https://forum.modular.com/t/parquet-mojo-available-for-your-columnar-data-needs/3470/2 "2026-09-08T02:22:40Z")

</div>

Iceberg.mojo and transitively parquet.mojo are now evaluated against pyiceberg on the well-known taxibench: [GitHub - magmalake/taxibench.example: The NYC taxi query suite over Apache Iceberg, run twice — once on iceberg-mojo, once on PyIceberg — and packaged as two container images. · GitHub](https://github.com/magmalake/taxibench.example)

Performance is comparable, the mojo implementation is slightly faster but I wouldn’t read too much in it. The new codebase may be missing more (adaptive) optimizations that make this workload faster at the expense of other use cases.