The next version of Microsoft’s open source machine learning tools arrived today to give developers a hand in creating intelligent systems. The Microsoft Cognitive Toolkit (previously known as CNTK) provides a set of tools to help developers build systems based on deep learning, without requiring PhD-level knowledge.

This revision of the Cognitive Toolkit provides a number of new features, including beta support for Keras, a popular high-level Python API for quickly coding up neural networks. TensorFlow and Theano, two other machine learning frameworks, already support Keras, and this update means it would be possible for data scientists to easily port their code between three different backends.

The toolkit also includes support for compressing models to run on lower-powered devices, making it easier for companies to roll out machine learning to edge devices.

Thursday’s updates build on the version 2 beta release process that Microsoft began last year. The beta included a number of key additions, including support for Python.

One of the key benefits Microsoft’s tools offer is their pedigree. The company uses its Cognitive Toolkit to create intelligent applications underpinning things like Bing video search.

“Unlike many toolkits, our toolkit — I want to really just highlight — is built to work reliably with massive production data out of the box,” Microsoft technical fellow Xuedong Huang said in an interview. “Most AI workloads within Microsoft, from Cortana to Bing to the Emotion API in Cognitive Services, they were all created using the Cognitive Toolkit.”

A recent benchmark by a team at Hong Kong Baptist University showed neural networks created with an earlier beta of Congnitive Toolkit 2.0 running faster than other competing frameworks on a wide variety of GPU-accelerated machine learning tasks.

The Cognitive Toolkit isn’t the only game in town, however. Microsoft is competing for data scientist and developer mindshare with a bunch of other programming frameworks backed by competing tech titans. Google has TensorFlow, which is the most popular open source machine learning framework on GitHub. Amazon Web Services has thrown its weight behind Apache MXNet, and Facebook is pushing its own Caffe2 framework.

All of these tools are part of a broader trend to provide developers with the tools to take advantage of machine learning at a time when it’s becoming an increasingly popular way to improve applications. Microsoft and its competitors are offering a wide variety of tools — from low-level frameworks all the way up to pre-built machine learning models — that developers can plug into an application without any training.