Installation and Requirements#


If you are reading this document on the Python Package Index (PyPI,, it is specific to the version of gevent that you are viewing. If you are viewing this document on, it refers to the current state of gevent in source control (git master).

Supported Platforms#

This version of gevent runs on Python 3.8 and up, (for exact details of tested versions, see the classifiers on the PyPI page or in gevent requires the greenlet library and will install the cffi library by default on Windows. The cffi library will become the default on all platforms in a future release of gevent.

This version of gevent is also tested on on PyPy 3.10 (7.3.12); it should run on PyPy 3.9 and above. On PyPy, there are no external dependencies.

gevent is tested on Windows, macOS, and Linux, and should run on most other Unix-like operating systems (e.g., FreeBSD, Solaris, etc.)


Windows is supported as a tier 2, “best effort,” platform. It is suitable for development, but not recommended for production. In particular, PyPy3 on Windows may have issues, especially with subprocesses.

On Windows using the deprecated libev backend, gevent is limited to a maximum of 1024 open sockets due to limitations in libev. This limitation should not exist with the default libuv backend.

Older Versions of Python#

Users of older versions of Python 2 or Python 3 may install an older version of gevent. Note that these versions are generally not supported.

Python Version

Gevent Version





<= 2.7.8




3.4.0 - 3.4.2






2.7.9 - 2.7.18, 3.6, 3.7




This section is about installing released versions of gevent as distributed on the Python Package Index. For building gevent from source, including customizing the build and embedded libraries, see Installing From Source.

gevent and greenlet can both be installed with pip, e.g., pip install gevent. Installation using buildout is also supported.

On Windows, macOS, and Linux, both gevent and greenlet are distributed as binary wheels.


You need Pip 8.0 or later, or buildout 2.10.0 to install the binary wheels on Windows or macOS. On Linux, you’ll need pip 19 to install the manylinux2010 wheels.


While the x86-64 binaries are considered production quality, they are built with relatively low optimization levels and no hardware specific optimizations. Serious production users are encouraged to install from source with appropriate compiler flags.


Beginning with gevent 20.12.0, 64-bit ARM binaries are distributed on PyPI for aarch64 manylinux2014 compatible systems. Installing these needs a very recent version of pip. These wheels do not contain the c-ares resolver, are not tested, and are built with very low levels of optimizations. Serious production users of gevent on 64-bit ARM systems are encouraged to build their own binary wheels.

Beginning with gevent 22.10.0, ppc64le binaries are distributed on PyPI. The same caveats apply as for 64-bit ARM binaries. Using them for anything other than local development is discouraged.

Beginning with gevent 23, muslinux aarch64 and S390X binaries are distributed on PyPI. The same caveats apply as for 64-bit ARM binaries. Using them for anything other than local development is discouraged.

Installing From Source#

If you are unable to use the binary wheels (for platforms where no pre-built wheels are available or if wheel installation is disabled), you can build gevent from source. A normal pip install will fall back to doing this if no binary wheel is available. See Installing From Source for more, including common installation issues.

Extra Dependencies#

There are a number of additional libraries that extend gevent’s functionality and will be used if they are available. All of these may be installed using setuptools extras, as named below, e.g., pip install gevent[events].


In versions of gevent up to and including 20.5.0, this provided configurable event support using zope.event and was highly recommended.

In versions after that, this extra is empty and does nothing. It will be removed in gevent 21.0.


Enables a pure-Python resolver, backed by dnspython. On Python 2, this also includes idna. They can be installed with the dnspython extra.


This is not compatible with Python 3.10+ or dnspython 2.


Enhancements to gevent’s self-monitoring capabilities. This includes the psutil library which is needed to monitor memory usage. (Note that this may not build on all platforms.)


A shortcut for installing suggested extras together. This includes the non-test extras defined here, plus additions that improve gevent’s operation on certain platforms (for example, in the past, it has included backports of newer APIs).


Everything needed to run the complete gevent test suite.