Put z = 3 into 1/(1+exp(-3)) and the workbench answers 0.95257412682243. At z = 1 the same expression gives 0.73105857863. At z = 0 the sigmoid is exactly one half, and the display shows it as the fraction 1/2 — the midpoint of the curve is not a rounded decimal.
Going backwards is worth typing too: to reach 0.95 you need z = ln(0.95/0.05) = 2.9444389791664, just under three, which is why 'three log-odds units' is the rule of thumb for a confident prediction. Feed that z back into the sigmoid and the screen hands it over as the exact fraction 19/20 — 0.95 and nothing else.
Answer
0.95257412682243
How to read it
1/(1+exp(-3)). The exact 95% point sits at z = ln(0.95/0.05) = 2.9444389791664, which returns the fraction 19/20.
Source
The logistic function from Verhulst's population model (1838); the same curve is the link function of logistic regression in any statistics or machine-learning text
Work it in the calculator
These links open the right calculator with this problem's numbers already typed in. The line above is what the screen should show — check it yourself.