The 0.12.0 JIT compiled hot loops on ARM64. As of 0.13.0 it does the same on x86-64, on macOS and Linux, which between them covers Apple silicon, the Intel Macs, and effectively every server you would deploy to.
Same rules on both architectures: opt in with --jit=on,
hot loops compile to native code while the rest interprets, and the
output stays byte-for-byte identical to the interpreter. The
differential test suite that enforces that identity now runs every
program on both instruction sets, which has the pleasant side effect of
making each backend a correctness check on the other.
Platforms that forbid runtime code generation, the consoles and iOS, are served by the interpreter, and the JIT compiles out of those builds entirely. Making that interpreter fast in its own right is an ongoing project we will have more to say about.