The math.floats.small.c-types vocabulary defines two-byte half and bfloat memory element types. They can be used with alien memory accessors. On ARM64, scalar calls by value use the C _Float16 and __bf16 conventions; direct calls, indirect calls and callbacks are supported when the C toolchain supports these types. Values convert numerically, using round-to-nearest ties-to-even and canonical NaNs. Windows ARM64 variadic signatures pass named and anonymous scalar half/BF16 payloads through the general-purpose argument registers and stack. Other architectures retain memory-only support.