If you use Python and you haven’t heard of Astral uv it might be time to take a look. uv claims to be:

A single tool to replace pip, pip-tools, pipx, poetry, pyenv, twine, virtualenv, and more.

Great! I use pip, pyenv, and virtualenv. It also claims to be:

10-100x faster than pip.

Wow. Big if true.

Let’s do our own real-world benchmark

Testing uv

Change is hard and changing software tooling is no exception

There can be a real productivity hit that needs to be balanced when learning new tools and technology. In addition, it’s not every day that we need to install a new version of Python with pyenv or set up a virtualenv for a new project

Now that I think about it, I don’t tend to think of pip install as a bottleneck in my workflow. Would a 10 - 100x speed up even matter to me?

The test

I took a Docker image that runs Streamlit and all of the related data analysis packages you can think of for my test. This particular image is built with docker buildx for multi-platform support: building an linux/arm64 image on linux/amd64

A single build of the image took around 6 minutes and 9 seconds without any layers or caching or other clever solutions

When changing the base image from:

python:3.11-slim

to:

ghcr.io/astral-sh/uv:python3.11-bookworm-slim

and changing the pip install command from:

pip install -r requirements.txt

to:

uv pip install -r requirements.txt --system

the same build took just 1 minute and 33 seconds, a 75% reduction! That’s a win-win-win: reduced development and deployment time, reduced cloud bills, and reduced environmental impact

Let’s visualize this comparison:

Conclusion

While we didn’t find that uv pip was 10 - 100x faster than pip, we did see a 75% reduction in Docker image build time from a 2 line code change – essentially for free

Thanks Astral team! I look forward to playing with uv more :)