Both answers tanh(C1-t) and 1/tanh(C1-t) are equivalent for different values of the constant. It's important to understand that the constants can be non-real:
In [48]: C1, C2 = symbols('C1, C2')
In [49]: sol1 = 1/tanh(C1 - t)
In [50]: sol2 = tanh(C2 - t)
In [51]: sol1.subs(C1, C2 - I*pi/2).rewrite(tanh) == sol2
Out[51]: True
Of course ideally the answer should be given in a form such that for real initial conditions the constants are real-valued. In this particular case it isn't possible to do that for all possible choices of real initial conditions:
If the initial condition is in the range (-1, 1) then the tanh(C - t) form can satisfy them with real C. However if the initial condition is greater than 1 or less than -1 then only the 1/tanh(C-t) form will satisfy the initial condition with real C. Neither form of the solution is preferable for all possible choices of initial conditions.
What is an issue is the fact that if you give initial conditions then dsolve errors out:
In [59]: dsolve(eq, B(t), ics={B(0):0})
...
NotImplementedError: Initial conditions produced too many solutions for constants
This happens because solving for the constants gives:
[{C1: -I*pi/2}, {C1: I*pi/2}]
Either choice of value for the constant gives the correct solution though:
In [11]: sol = dsolve(eq, B(t))
In [12]: print(sol)
Eq(B(t), 1/tanh(C1 - t))
In [13]: print(sol.subs(C1, -I*pi/2))
Eq(B(t), -1/coth(t))
In [14]: print(sol.subs(C1, I*pi/2))
Eq(B(t), -1/coth(t))
The initial condition handling code in dsolve should be fixed to handle cases like this.
Oscar