We follow the Peng/Donoho/Clarebout definitions and consider reproducibility to be same data + same methods = same results, while replication involves new data and isomorphic methods. Rokem, Marwick, and Staneva provide a nice overview of debate around term use in their chapter in The Practice of Reproducible Research.
More specifically, Code Ocean is (first and foremost) a platform for computational reproducibility, what Kitzes calls the ability to “recreate the final reported results of the project, including key quantitative findings, tables, and figures, given only a set of files and written instructions.” Stodden distinguishes computational from statistical and empirical reproducibility, all three of which are complementary; as Zelner notes, computational reproducibility, the ability to recreate “analyses using the exact same input data ... is a necessary precondition to the new-data replication.”
Code Ocean also facilitates replication. All published capsules will accept new data and changes to the code (which will not affect the publicly accessible version, but rather create a private edition whose edits only you can see). Authors For an example, see this published capsule, which you can test on an input video of your own.
For a thorough overview, see Barba (2018) Terminologies for Reproducible Research.