Limited precision of float numbers is deterministic. But there's whole parallelism and how things are wired together, your generation may end up on a different hardware etc.

And models I work with (claude,gemini etc) have the temperature parameter when you are using API.