It's a tie between Julia and Nim for me. Both have a high-level, readable syntax while also being natively very very fast.
Julia is great for exploratory numerical/scientific computing, e.g., AI, simulations, etc. It especially has amazing math syntax and unicode character support, making for really elegant math code.
Nim is a systems programming language, and I've been starting using it for embedded systems lately. I think it could be really good for running machine learning on embedded devices, as C/C++ are kinda miserable for that, but MicroPython is way too slow and not well suited for production embedded systems imo. Plus it compiles to C and C++, so you can compile it and run it on any device for which you have a working C or C++ toolchain.