Light. Only the essentials are included, with six and mimeparse the only dependencies beyond the standard library. We work to keep the code lean, making Falcon easier to test, optimize, and deploy.
Fast. Cloud APIs need to turn around requests quickly, and make efficient use of hardware. Falcon processes requests several times faster than other popular web frameworks.
Flexible. Falcon can be deployed in a variety of ways, depending on your needs. The framework speaks WSGI, and works great with Python 2.6 and 2.7, PyPy, and Python 3.3/3.4. There's no tight coupling with any async framework, leaving you free to mix-and-match what you need for your web service.
Performance. Unlike other Python web frameworks, Falcon won't bottleneck your API's performance under highly concurrent workloads. Many frameworks max out at serving simple "hello world" requests at a few thousand req/sec, while Falcon can easily serve many more on the same hardware.
Freedom. Falcon isn't very opinionated. In other words, the framework leaves a lot of decisions and implementation details to you, the API developer, which means you will need to use your head a little more than other web frameworks, and probably write a little more code. On the other hand, this gives you a lot of freedom to customize and tune your implementation in order to create a solution that stands out from the crowd.
Reliability. Falcon doesn't include a lot of cruft that is simply unnecessary when building web services. Less code and fewer dependencies means a smaller attack surface, lower memory usage, and fewer places for bugs to hide. Our build fails at anything less than 100% code coverage. We use Travis and Coveralls to keep ourselves honest.
If you're still not convinced, check out this bunny with a pancake on its head.
Scenario. The benchmark acts as a WSGI server and performs a GET request directly on each framework's app. Each app parses a route template with a single embedded parameter, reads a query parameter and a header from the request data, sets an x-header on the response, and finally returns a 10 KiB plain-text body, randomly generated.
Method. 200,000 iterations were executed per trial, and the best time was recorded for each framework over 20 trials. The order in which the frameworks were tested was randomized for each trial. Also, garbage collection was enabled as it would be in a production environment, and a full collection was forcefully triggered before each trial. Finally, results were calculated using the Decimal class, and rounded to the nearest whole number.
Platform. Benchmarks were executed on a Rackspace Cloud Server (1 GB Performance), running Arch 2013.9 with all system packages updated to their latest versions as of 16 Jan 2014. The frameworks were tested on several different versions of Python. Note that in these tests the Falcon modules were not cythonized; generally speaking, the reader may expect a ~10% boost in performance when Falcon is installed with Cython.
Extended Test. Falcon was also benchmarked under Python 2.7.6 with a more realistic scenario, in which the routing table had multiple entries, the query string contained percent-encoded characters, and several complex response headers were set in the response. Note that even when Falcon is doing more work, it still is able to outperform other frameworks, such as Bottle, which are only doing the bare minimum to construct a response.