.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
Short answer, as of August 2026: Miniconda is free for anyone to download and install, and the conda tool inside it is BSD-licensed open source. What Anaconda sells is access to its package repository, the defaults channels served from repo.anaconda.com, and organizations with more than 200 employees need a paid license to pull from there. Most of the confused forum threads about "the Miniconda license" come from mixing those pieces together, so it's worth pulling them apart properly before getting to the practical fixes.
What's free and what's licensed in Miniconda
The Miniconda installer is a free download covered by Anaconda's end user license agreement. conda itself is open source under a BSD license, usable, forkable and redistributable at any company size. The package repository is the commercial piece: the curated defaults channels Anaconda builds, hosts and licenses. Anaconda Distribution bundles that same repository access with a 600-package head start, a trade I've weighed separately in Miniconda vs Anaconda, and the licensing question lives entirely in the repository piece, for both products equally.
The 200 employee threshold, counted org-wide
Anaconda's terms of service FAQ draws the line at 200 total employees or contractors, counted across the entire organization including all affiliates, an affiliate being any entity under more than 50% common ownership or control. So a 150-person subsidiary of a 5,000-person group is over the line, and contractors count toward it too. The terms carve out free use for individuals doing personal non-commercial work, for eligible academic institutions and for eligible nonprofit and research organizations.
What changed between 2020 and 2024
The 2020 terms update was the first to put commercial repository use behind a paid license, and it kept catching teams by surprise for years afterwards. The 2024 revisions then tightened the counting to the whole organization with affiliates included, and narrowed the research and nonprofit exemptions many universities had assumed covered them; course-based education later got its free access restored. That's the state of things as of August 2026. Given that the terms have moved twice in six years, re-read the source before any purchasing decision rather than trusting this paragraph's age.
The .condarc trap in Miniconda 24.9.2 and later
Since version 24.9.2 the Miniconda installer writes a .condarc file that configures Anaconda's channels explicitly, and this is where careful people get caught. A default install pulls from the commercially licensed repository without anyone having chosen a channel, so the very first conda install numpy at a 1,000-person company already falls under the terms of service. The installer being free changes nothing about what the installed tool then downloads.
Check which channels you're using
conda config --show channelsOutput including defaults means package operations resolve against repo.anaconda.com and the licensing question applies to you; a list showing only conda-forge or other community channels means it doesn't. Since .condarc files are per-install and per-user, this check belongs on every machine and CI runner, and in audits I've been near, the forgotten build server still on defaults is the finding every single time.
How to switch to conda-forge
conda-forge is the community packaging channel, free at any organization size. Moving an existing Miniconda install onto it takes three commands, the first two straight from the conda-forge documentation:
conda config --add channels conda-forge
conda config --set channel_priority strict
conda config --remove channels defaultsThe third removes Anaconda's channels so nothing falls through to them, after which new environments resolve purely against conda-forge. What remains is the base environment, whose Python and conda came from defaults at install time; most teams accept that leftover and move on, since nothing new arrives from the licensed repo. Make the switch before updating Miniconda, so the update itself already resolves against conda-forge.
