I like the suggestion of using both! 
And of course, you’re right about benchmarks not being like for like. However, they are still benchmarks and the discrepancy between the encode and decode performance is interesting to think about.
1 Like
