It's the extended type. But there is currently a Double precision limit on literal numbers.
All floating point calcs are done in extended precision on the FPU, even if the variables are singles. The precision is only limited when the value is stored to a variable or pushed onto the stack.
Charles