Skip to content

Miniconda

Anaconda, Inc. · Minimal conda and Python installer for isolated project environments and command-line package management.

CaskLanguages & RuntimesAuto-updates 7.6K
Install command
brew install --cask miniconda
30-day installs
2,545
#129 in App
1-year installs
62.4K
Last 365 days
Latest version
py314_26.9.4-1
Download size
176MB
sh
Release cadence
Weekly
3 releases in 30 days

About

Miniconda is Anaconda's minimal installer for conda, Python and the dependencies needed to run them. It provides a command-line starting point for developers, data scientists and automation users who want to install only the packages each project needs. It does not include Anaconda Navigator or the broad preinstalled data-science collection of Anaconda Distribution.

Main features

  • Create separate environments with different Python and package versions.
  • Install, update and remove packages with dependency resolution through conda.
  • Activate project environments without replacing the macOS system Python.
  • Choose package channels and manage repository access according to their terms.
  • Export and recreate environment specifications for development and automation.
  • Use a compact base installation, then add scientific, development or other packages as needed.

Installation and first use

Install this Mac cask with brew install --cask miniconda. It installs into Homebrew's Caskroom base directory and links the conda command. Follow the cask's shell-setup guidance; for zsh, run conda init zsh, then reopen Terminal. Check installation with conda list. For example, create a separate project environment with conda create -n myproject python=3.14, activate it using conda activate myproject, and install the packages your project requires. These are usage instructions, not operations performed by this catalog.

The official shell or graphical installer is an alternative to Homebrew, with a different default installation location. Avoid confusing an existing Anaconda, Miniforge or Miniconda prefix with the cask-managed installation. Review and accept the relevant installer and repository terms. Keep project environments separate from base. Package downloads generally require network access; optional private repositories require their own credentials.

Requirements, dependencies and conflicts

Current official Miniconda installers support Apple Silicon Macs running macOS 12.1 or later. The recorded Apple Silicon cask version py314_26.7.1-1 selects the Python 3.14 installer containing Python 3.14.7 and conda 26.7.1. Its ARM64 shell installer is 155,681,303 bytes; this is download size, not installed disk use.

Anaconda ended new Intel macOS installer releases after the 25.7.x line. Homebrew's Intel variation therefore remains py313_25.7.0-2, rather than matching the current Apple Silicon version. Do not assume the latest ARM64 release is an Intel-native download. Older archived installers have their own compatibility limits.

The Homebrew cask explicitly conflicts with miniforge and declares no additional formula dependency. Its raw macOS dependency field does not establish the vendor's current OS support policy. Official system requirements list a minimum 400 MB to download and install; the current comparison page estimates roughly 900 MB for an installed Miniconda base, and environments/packages need additional space. No universal RAM requirement was established.

Official non-Mac installers cover Windows 64-bit x86 and Linux x86-64/aarch64. Current Linux installers require glibc 2.28 or newer; aarch64 builds target server-class hardware and may not suit every Raspberry Pi. Consult the current system-requirements page for Windows and distribution support.

Accounts, fees and licensing

The Miniconda installer is available without a separate installer purchase, but its EULA and package licenses apply. Conda itself is an open-source component; that does not make every bundled package or repository service subject to the same license. Access to Anaconda's repositories/platform can require a paid subscription. Current terms require a Business Plan for use on behalf of a for-profit organization with more than 200 employees or contractors, including affiliates, unless an applicable exception qualifies it for free use. Personal non-commercial use, qualifying academic/non-profit use and eligible smaller businesses have defined free-use provisions. Review the full current terms and organizational eligibility before choosing channels.

An account is not inherently required to create and use local conda environments. Anaconda's web download flow asks for registration, and authenticated platform/private-repository features need appropriate credentials. The installer EULA explicitly prohibits bypassing repository payment requirements.

The icon is Anaconda's official publisher brand symbol. The media example comes from official Miniconda Windows installation documentation: it shows opening Anaconda Prompt on Windows 11, not a macOS interface or a GUI bundled with the Mac cask.

Sources: Miniconda overview, System requirements, Installation, Official installer archive, Miniconda EULA, Anaconda terms, Conda documentation.

At a glance

Chip
Apple silicon
Auto-updates
Yes
Repository
Open source · BSD-3-Clause · Updated 1 day ago★7.6K
Developer
Anaconda, Inc.
Homebrew
Available; the app updates itself
Also known as
miniconda
Homepage
www.anaconda.com/docs/getting-started/miniconda/main
Token
miniconda

Checked

What’s new in 26.10.0

New · From GitHub Releases

✨ Special announcement ✨

This is an announcement about an important change in conda's functionality:

With this 26.10.0 release we are changing the default solver of conda to conda-rattler-solver! 🥳 🚀

The previous default, the libmamba solver, remains available as an alternative, as does the pycosat solver (also known as classic). The classic solver now ships as the separate conda-pycosat-solver plugin, which is bundled with conda. Existing solver: classic settings and --solver=classic commands keep working, and the plugin also accepts --solver=pycosat.

Why are we switching the solver?

In short: to make conda faster and more accurate.

Rattler is a set of Rust libraries for working with conda packages. The conda-rattler-solver plugin uses rattler's dependency solver and optimized sparse repodata parsing, which reduces overhead when resolving complex dependency trees. It solves environments faster than libmamba, with the largest improvements on large or heavily constrained environments.

Conda 26.10.0 switches to rattler for fresh installations and upgrades that do not have an explicit solver setting. If you have already selected a solver, conda keeps that choice.

What can I do if this update doesn't work for me?

If the new solver is not working as you expect:

  • Check if the behavior you are observing is a known limitation or a deliberate change.
  • If that's not the case, please submit a bug report or feature request in the conda-rattler-solver repository.
  • If necessary, you can go back to using the libmamba solver without modifying your conda installation:
    • When possible, pass the command line option --solver=libmamba to your conda calls.
    • Otherwise (e.g. for conda build ... or constructor ...), set the environment variable CONDA_SOLVER=libmamba.
    • For permanent changes, use the conda configuration system: conda config --set solver libmamba.
  • The pycosat solver (also known as classic) remains available with --solver=pycosat or --solver=classic.

Where can I learn more about conda-rattler-solver?

The documentation of the conda-rattler-solver plugin can be found at conda.github.io/conda-rattler-solver, including the known limitations and differences, and in the conda solvers guide.

Enhancements

  • IMPORTANT: Set solver: rattler as the new default solver, including after upgrades when no solver is explicitly configured. Require conda-rattler-solver >=26.10.2. (#16853 via #16944)
  • Track the last time an environment was activated via user invocation. Activation modifies the mtime of <prefix>/conda-meta/last_activated (best-effort; failures are ignored). The timestamp is exposed as last_activated in JSON output. (#14308 via #16808)
  • Add an is_active property to PrefixData. (#14308 via #16808)
  • Add --disable-plugins to selectively disable plugins at runtime using comma-separated names or by repeating the option. Keep --no-plugins as a boolean flag to disable all external plugins. (#15914 via #15915)
  • Add --enable-plugins to keep selected plugins enabled when using --no-plugins or CONDA_NO_PLUGINS, including solver plugins needed for recovery. (#15914 via #15915)
  • Add a root_extras option to conda._private.shards.subset.build_repodata_subset (and the underlying RepodataSubset.reachable* methods) so callers can request that CEP 44 extra_depends groups be fetched only for specific root/user-requested packages, without recursing into extras of their dependencies. (#16724 via #16742)
  • Add LinkSourceNotFoundError and LinkScriptError (a LinkError subclass) so callers can detect a missing package cache source and a failed pre-/post-link script with isinstance and inspect src, action, prec, and path instead of parsing error messages. Error text is unchanged, and LinkScriptError remains catchable as LinkError, but code that matched the missing-source case against a bare CondaError now needs LinkSourceNotFoundError. (#16785 via #16747)
  • Follow CEP 44 extras explicitly requested by dependency specs (e.g. pkg[extras=more]) when fetching sharded repodata, and revisit a package when another dependency activates an additional extras group. Groups that were not requested stay inactive. Previously only extras requested for root packages via root_extras were followed. (#16891 via #16908)
  • Parse MatchSpec strings with the v3 parser for all solvers, not just for rattler. This keeps complete extras and flags lists (e.g. httpx2[extras=[cli,http2]] or httpx2[extras=cli,http2]) and accepts nested when expressions. (#16724 via #16906)

Bug fixes

  • Preserve platform-named parent directories when parsing channel and package URLs, fixing conda-build failures with paths such as osx-arm64/conda-bld. (conda/conda-libmamba-solver#1108 via #16926)
  • Preserve nested shard URLs when adding stored Anaconda tokens. (#16946 via #16948)
  • Avoid FileNotFoundError on Windows when rmtree() deletes a directory that RD /S /Q already removed (for example during conda package temp cleanup). (#16539 via #16803)
  • Write configuration files atomically when directory permissions allow, preserve macOS ACLs, and reject writes if the file changed in the meantime. Keep writable files editable in read-only directories, with best-effort recovery after write failures. (#16694 via #16695)
  • Ask once before conda env remove and conda remove --all delete an environment. Those commands previously confirmed the package unlink and then, after the packages were already uninstalled, asked again before deleting the rest of the prefix, including non-conda files. Declining that second prompt left a half-deleted environment. The package plan and the prefix-deletion warning are now shown together before anything is removed, and declining leaves the environment unchanged. (#16753 via #16852)
  • Fix conda config --get writing to the config file when no key is given. (#16825 via #16831)
  • Fix iter_records_v3() not converting raw-bytes hashes (sha256, md5, legacy_bz2_md5, attestations_sha256) to hex strings, as written by e.g. rattler-index. (#16866)
  • Retry remote requests that fail with HTTP 502 (Bad Gateway) or 504 (Gateway Timeout), which are usually transient server or CDN errors. (#16848)
  • Fix ObjectParameter configuration types not expanding environment variables when expandvars=True. (#16897)
  • Fix conda.deprecations.deprecated and conda.deprecations.deprecated.argument to preserve generator functions when decorating, instead of wrapping them in a plain function and losing generator semantics (e.g. pytest yield-fixtures now warn and yield correctly when stacked under @pytest.fixture). (#16900)
  • Use a fresh YAML parser for every read and write to prevent concurrent calls and failed operations from corrupting later results. (#16943 via #16949)
  • Show environment names in PowerShell prompts for canary packages built with changeps1: false. (#16858)

Deprecations

  • Mark conda.plugins.solvers, conda.resolve, conda.common.logic, and conda.common._logic as pending deprecation, to be removed in 27.9. Use conda-pycosat-solver instead. (#14170 via #16520)
  • Mark conda.models.prefix_graph.GeneralGraph and its breadth_first_search_by_name method as pending deprecation, to be removed in 27.9. (#16802)
  • Mark conda.testing.cases as pending deprecation, to be removed in 27.9. Use native pytest fixtures instead. (#16900)
  • Mark conda.testing.helpers.assert_equals, assert_in, and assert_not_in as pending deprecation, to be removed in 27.9. Use a plain assert instead. (#16900)
  • Mark conda.testing.helpers.supplement_index_with_repodata as pending deprecation, to be removed in 27.9. Use conda.models.records.PackageRecord.from_objects instead. (#16900)
  • Mark conda.testing.helpers.add_feature_records_legacy as pending deprecation, to be removed in 27.9. (#16900)
  • Mark conda.testing.integration.Commands as pending deprecation, to be removed in 27.9. Use the conda_cli fixture instead. (#16900)
  • Mark conda.testing.integration.escape_for_winpath as pending deprecation, to be removed in 27.9. Use re.escape instead. (#16900)
  • Mark the conda.testing.integration constants UNICODE_CHARACTERS, UNICODE_CHARACTERS_RESTRICTED, SPACER_CHARACTER, and cp_or_copy as pending deprecation, to be removed in 27.9. (#16900)
  • Mark conda.testing.solver_helpers.temp_simple_env as pending deprecation, to be removed in 27.9. Use conda.testing.solver_helpers.SimpleEnvironment instead. (#16900)
  • Mark conda.models.match_spec._parse_spec_str as pending deprecation, to be removed in 27.9. Use conda.models.match_spec._parse_spec_str_v3 instead. (#16724 via #16906)
  • Mark the unused conda.core.subdir_data.SubdirData pickle cache helpers and the cache_path_pickle property as pending deprecation, to be removed in 27.9. (#16928)

Docs

  • Document the last_activated timestamp in the JSON output. (#14308 via #16808)
  • Add a "Rendering output" deep dive documenting the reporter backend as the output system for the conda CLI, with logging and error handling guidelines. (#16700 via #16843)
  • Document logging guidelines for library code in the logging deep dive. (#16701 via #16912)
  • Add a guide to choosing a solver to the documentation, with information about the available solvers and how to change your solver. (#16844)

Other

  • Move the classic solver out of conda core and into the standalone conda-pycosat-solver plugin, bundled with conda by default. Both --solver=pycosat and --solver=classic continue to work unchanged. (#14170 via #16520)
  • Keep PowerShell CI installations separate from PowerShell's runtime cache. (#16762 via #16907)
  • Add tmp_repodata_channel for constructing local test channels from in-memory package metadata. (#16807)
  • Replace production-channel inputs in test_neutering_of_historic_specs with a small local channel while preserving historical-spec regression coverage. (#16767 via #16807)
  • Speed up test_cross_channel_incompatibility by using two small local channels instead of remote conda-forge/defaults metadata, preserving optional-constraint regression coverage. (#16768 via #16796)
  • Raise the minimum Python version required to run conda to 3.11, following conda-build#6176. Conda can still create and manage environments containing Python 3.10 and older; this only affects the interpreter used to run conda itself. (#16795 via #16801)
  • Do not allow local condarc files to populate the context_testdata fixture. (#16797 via #16809)
  • Warn when a conda doctor reinstall attempt fails due to malformed conda-meta data. (#16636 via #16834)
  • Convert raw-bytes hash digests to hex strings in memory when sharded repodata is decoded, rather than re-converting records on each read. (#16868)

Contributors

Full Changelog: https://github.com/conda/conda/compare/26.9.3...26.10.0

How to install Miniconda on Mac

Install without Homebrew

Download the same file Homebrew uses: Install Details

Uninstall Miniconda and optional data cleanup

Uninstall command
brew uninstall --cask miniconda

A regular uninstall removes the app and keeps its settings and caches.

Deep clean (optional)
Uninstall command
brew uninstall --zap --cask miniconda

--zap also deletes the settings, caches and other files listed in the Homebrew definition. They can include data shared with other apps and cannot be restored, and it does not guarantee that nothing is left. Back up any data you need first. To keep settings and caches, leave out --zap.

Questions about Miniconda

Is Miniconda safe?

Miniconda is in Homebrew’s official catalog and has not been disabled. Since September 2026, Homebrew disables apps that fail macOS Gatekeeper checks. When installing, Homebrew checks the download from repo.anaconda.com against its SHA-256 checksum. It was installed 2,545 times through Homebrew in the last 30 days.

Is Miniconda open source?

Miniconda’s source code is public on GitHub (conda/conda) under the BSD-3-Clause license.

How do I completely uninstall Miniconda on a Mac?

Run brew uninstall --cask miniconda. To delete its settings and caches as well, run brew uninstall --zap --cask miniconda. --zap also deletes the settings, caches and other files listed in the Homebrew definition. They can include data shared with other apps and cannot be restored, and it does not guarantee that nothing is left. Back up any data you need first. To keep settings and caches, leave out --zap. That also deletes 4 files and folders, including ~/.conda.

Where does Miniconda store its settings and data?

Its data is in ~/.conda, ~/.condarc, and ~/.continuum. These locations come from Homebrew’s clean-up list.

What is the latest version of Miniconda?

The latest version is py314_26.9.4-1, added to Homebrew on Nov 10, 2026. Miniconda updates itself; to update it with Homebrew, run brew upgrade --cask --greedy miniconda.

What else does installing Miniconda add?

It also adds 1 command-line tool: conda. It cannot be installed alongside miniforge.

Similar Apps

  • MiniforgeMinimal Conda and Mamba installer preconfigured for conda-forge, with isolated package environments and native multi-platform installers.
  • Eclipse Temurin 25Free OpenJDK-based Java 25 development kit with the HotSpot JVM, compiler and diagnostic tools.
  • Azul Zulu 17Azul’s OpenJDK Java 17 development kit for running, compiling and maintaining Java applications.
  • Eclipse Temurin 8Eclipse Adoptium’s OpenJDK Java 8 JDK with HotSpot and development tools; this Mac package is Intel x64 and requires Rosetta on Apple Silicon.
Browse all apps