Arguably, it was the second.

The Harvard Mark 1 ran its first program in 1944, but didn’t have branches as we understand them until 1946:

https://en.wikipedia.org/wiki/Harvard_Mark_I

Apparently, you could achieve loops by taping the input program into a physical loop, even in 1944.

Colossus Mark I beat them too:

https://en.wikipedia.org/wiki/Colossus_computer