For most cases, especially for OpenVINO applications, use "Python3 (Ubuntu)". This version comes from the OS on the development node and it will most closely match your experience in jobs on compute nodes.
The kernel "Python3" comes by default with the Jupyter Notebook and it is the kernel used to run the Jupyter Server interface itself. There is little reason to use that for DevCloud workloads.
Finally, in a couple of weeks, you may see a new option, "Intel Distribution for Python". This is for special use cases, such as using TensorFlow, Keras, or other DL frameworks that use the MKL DNN functionality on Intel CPUs. For OpenVINO, "Python3 (Ubuntu)" will give you better compatibility.
This is with kernel "Python 3 (Ubuntu)", the preferred version:
This is with kernel "Python 3", do not use: