.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
Miniconda gets you a working Python and the conda package manager from one small installer, with nothing else riding along: none of the bundled libraries and no graphical launcher, the two things that make the full Anaconda download so heavy. You start from a clean base and build exactly the environments your projects need. I've run it for years on laptops, build servers and GPU boxes, and it's what I'd install today in nearly every case. This guide covers the whole surface, with a dedicated article behind each section for when you're ready to go deeper.
What is Miniconda
Miniconda is Anaconda's minimal installer for conda. One download drops a Python interpreter, the conda CLI and the handful of packages those two depend on into a single folder in your home directory, and that's the entire product. conda is the interesting half: a package and environment manager that installs prebuilt binaries, which matters enormously for the scientific stack, where compiling NumPy or a CUDA-linked library from source is nobody's idea of a good afternoon.
It also covers a gap pip and venv leave open. A conda environment carries its own Python, so one project can run 3.11 while another runs 3.14 without either touching the system interpreter, and conda packages ship non-Python dependencies too: C libraries, compilers, CUDA toolkits. For purely-Python projects venv remains fine, and I still use it for small ones. The moment binary dependencies or multiple Python versions enter the picture, conda becomes the better tool.
What Miniconda is not: the Anaconda Distribution, which preinstalls hundreds of packages and puts the Navigator GUI on top. There's no feature gap between them, since the conda inside both is identical. And Anaconda's installer isn't the only road to conda either, a point that returns when licensing comes up below.
The current Miniconda release
As I write this the current release is 26.5.3-2, published July 29, 2026, shipping conda 26.5.3. Anaconda builds installers against five Python versions for the base environment (3.10 through 3.14), plus "latest" filenames like Miniconda3-latest-Linux-x86_64.sh that always resolve to the newest build, which is the form provisioning scripts want since it never rots. Versioned filenames follow the pattern Miniconda3-py314_26.5.3-2-<platform> for pinning an exact build, and every installer sits on repo.anaconda.com/miniconda beside its SHA256 hash, which you should spend the ten seconds comparing before executing anything.
The Miniconda download guide untangles the full set of choices there, latest against pinned Python, x86_64 against ARM, .sh against graphical, since the index page itself explains none of them.
Installing on Linux, Windows and macOS
Linux is four commands: download the .sh installer with curl, verify with sha256sum, run it with bash, let it offer conda init at the end, and a new shell then greets you with (base). The installer's -b flag covers unattended installs, an aarch64 build covers ARM servers, and the step-by-step lives in the guide to install Miniconda on Ubuntu. When the target is a server, LumaDock's Ubuntu VPS plans carry a one-click Miniconda template that skips the manual setup altogether.
The Windows walkthrough is in the guide to install Miniconda on Windows, and its headline quirk is deliberate: conda stays off the system PATH, with the Anaconda Prompt (miniconda3) Start Menu entry as the supported way in, plus conda init powershell for wiring up a real terminal.
macOS gives you three routes, a .sh installer in the terminal, a graphical .pkg and the Homebrew cask (brew install --cask miniconda, followed by the conda init the cask reminds you about). My pick is the .sh on any Mac, partly for consistency with Linux muscle memory and partly because the .pkg has a history of refusing to run on some systems for no visible reason; picking the right architecture and the rest of the details are in the guide to install Miniconda on macOS.
