The actor paradigm has been proposed as a natural approach to the development of concurrent programs. It has lead to the development of several actor-based languages, each representing different points in the language design space. But how do these languages perform? The Savina collection of micro benchmarks attempts to approach this question, with multiple programs exercising different design aspects. In our work to understand actor language performance, we want to run as many of the Savina programs as possible in as many actor languages as possible on increasingly capable configurations. In this talk we compare three actor-based languages representing different ends in the design spectrum: Akka, CAF and Pony. We ran them on two architectures and using 2, 4, … 64 logical cores. We will show the results in beautiful plots. Be prepared for considerable disparity in the scaling behaviour across languages and programs. And what conclusions can we draw from this disparity? We will propose criteria describing the programs’ profiles and their performance profiles. We want to gauge your opinion as to how intuitive and predictive such profiles are, and what else they should include. . We will then attempt to correlate program profiles, scaling shapes with the implementation choices, again inviting your views on expected outcomes or co-relations to be considered.