Which Python package manager should I use?
Use uv. It handles dependency management, virtual environments, Python version management, and package building in a single tool. It follows Python packaging standards, generates cross-platform lockfiles, and resolves dependencies faster than any alternative. For most Python projects in 2026, uv is the right default.
Python has more package managers than most languages, so the decision tree below helps when uv does not fit.
Choose a package manager
graph TD
A[New Python project] --> B{Need conda packages?<br/>CUDA, GDAL, HDF5, etc.}
B -->|Yes| PIXI[pixi]
B -->|No| C{Existing project?}
C -->|No| UV[uv]
C -->|Yes| D{Current tool working?}
D -->|Yes| STAY[Keep current tool]
D -->|No| E{Currently using...}
E -->|Poetry| F{Poetry plugins<br/>with no uv equivalent?}
F -->|Yes| POETRY[Stay on Poetry]
F -->|No| MIGRATE_UV[Migrate to uv]
E -->|pip + requirements.txt| MIGRATE_UV2[Migrate to uv]
E -->|Pipenv| MIGRATE_UV3[Migrate to uv]
E -->|conda| G{Need conda-only<br/>packages?}
G -->|Yes| PIXI2[Migrate to pixi]
G -->|No| MIGRATE_UV4[Migrate to uv]
Check when uv is not enough
Install conda packages
Some dependencies do not exist on PyPI. CUDA toolkits, GDAL, HDF5, and other native libraries require conda channels. Pixi resolves conda and PyPI dependencies together in a single lockfile. It uses uv internally for PyPI resolution, so the workflow feels similar.
See When should I choose pixi over uv? and Why should I choose conda? for detailed comparisons.
Keep required Poetry plugins
Poetry has a plugin system that uv lacks. If your build depends on poetry-dynamic-versioning, a custom plugin for monorepo management, or another plugin with no equivalent, staying on Poetry avoids rework. Check whether uv covers your use case before deciding; many projects that used plugins find they no longer need them.
For projects without plugin dependencies, migrating from Poetry to uv is straightforward.
Keep a working setup
Migration has a cost. If your team uses pip-tools, hatch, or PDM and the workflow is stable, switching tools is not urgent. All three follow Python packaging standards, so your pyproject.toml will transfer when you do migrate. Hatch, PDM, and pixi can all use uv as their installer backend, so you can get uv’s speed benefits without changing your workflow.
Compare package managers
Choose uv
A single binary that replaces pip, pip-tools, pyenv, pipx, and virtualenv. It installs dependencies, creates virtual environments, pins Python versions, and builds packages. Cross-platform lockfiles and PEP 621 metadata keep your project portable across tools.
Choose uv for: new projects, teams standardizing on one tool, CI pipelines where speed matters.
Choose Poetry
Installs dependencies, creates virtual environments, builds packages, and publishes to PyPI. A plugin ecosystem and large user base set it apart from newer tools. Poetry supports standard project metadata and dependency groups, alongside its legacy [tool.poetry] tables.
Choose Poetry for: existing Poetry projects with plugin dependencies, teams already trained on Poetry’s workflow.
See How do uv and Poetry compare? for a detailed breakdown.
Choose pixi
Pulls from both conda and PyPI channels to resolve dependencies together. It installs native libraries, GPU toolkits, and non-Python software that PyPI cannot distribute. uv powers its PyPI resolution internally, and a built-in task runner replaces Makefiles or scripts.
Choose pixi for: data science with native dependencies, GPU/CUDA workflows, projects mixing Python with R or Julia.
See When should I choose pixi over uv?
Use pip with requirements.txt
The original Python package installer. No lockfile, no virtual environment management, no Python version management. Still works and still ships with Python. For scripts and quick experiments, pip install inside a venv remains a valid approach.
Choose pip for: throwaway scripts, environments where installing additional tools is restricted, legacy systems you cannot change.
If you want faster installs without changing your workflow, uv pip is a drop-in compatible interface that reads the same requirements.txt files. For a fuller migration, moving from requirements.txt to pyproject.toml with uv takes minutes.
Choose Hatch
A standards-compliant project manager with built-in environment management, PEP 751 lockfile generation, and a matrix testing system. Hatch’s build backend (hatchling) is widely used across the ecosystem. Hatch focuses on library development and publishing. It can use uv as its installer backend for faster dependency installation.
Choose hatch for: projects already using hatch, teams that want hatch’s environment matrix for multi-version testing.
Choose PDM
A standards-compliant project manager with lockfile support. PDM follows Python standards closely and can use uv as its installer backend. It also supports PEP 582 (__pypackages__ directory) as an opt-in feature, though PEP 582 itself was rejected by the Python Steering Council in 2023.
Choose PDM for: existing PDM projects with no pressing reason to migrate.
Keep an existing Pipenv workflow
An older tool combining pip and virtualenv with a Pipfile.lock. Development has slowed relative to uv and Poetry. New projects should choose uv instead.
Choose Pipenv for: existing Pipenv projects where migration is not yet practical. When it becomes practical, How to migrate from Pipenv to uv takes one command.
Weigh the trade-offs
Share standard metadata
Tools that use PEP 621 metadata ([project] table in pyproject.toml) make your project portable. uv, hatch, PDM, and flit all read the same format. Poetry also supports it, though existing [tool.poetry] metadata is not migrated automatically. Standard metadata transfers across tools; sources, plugins, and build settings still need review when switching.
Require reproducible installs
A lockfile pins every transitive dependency to an exact version and records hashes. uv, Poetry, PDM, pixi, and hatch all generate lockfiles. pip-tools generates requirements.txt files that serve a similar purpose. Plain pip has no lockfile mechanism.
For reproducible deployments and CI, lockfile support is non-negotiable.
Compare installation speed
The handbook recommends uv for faster dependency resolution and installation. Compare tools on your project’s dependencies and CI cache conditions before estimating savings.
That speed advantage extends beyond the uv CLI. Pixi, hatch, and PDM can all use uv as their installation backend, so even teams that stay on their current tool can benefit from uv’s resolver and installer.
Manage Python interpreters
uv, pixi, hatch, and PDM can all manage Python interpreters directly. Poetry also has experimental interpreter management, or can use an existing Python installation. Having one tool handle both the interpreter and dependencies reduces setup steps and avoids version mismatches.