In case of 'spark2' default interpreter, both '%spark' and '%spark2' point to Spark2.

Lets say you have a cluster with both spark1.6 and spark2.0 installed and zeppelin has 2 interpreters: spark and spark2

Create a notebook whose default interpreter is spark2.

Then no matter whether you run paragraphs with %spark or %spark2, both point to spark1.6 (try running sc.version with both and both will point to spark 1.6)