I've tested using a notebook with a cluster and runtime 15.3. 

I was able to create a table with variant data type using spark sql api