It is not mentioned anywhere that inputs like [0] * n, n > 1 and [n] + [0] * m, n != 0, m > 1 should be interepreted as geometric sequences.

For invalid sequences either the output should be changed to None, or the function should throw an error. Returning a magic value which never occurs as "normal" function output because your input range is artificially limited is plain bad design.

In some cases, which do not correspond to either AP or GP, first element of sequence is expected instead of -1

[198, 304, 540, 605, 961, 1165, 1325, 1596, 1635, 1900, 2123, 2318, 2503, 2699, 2814, 3051, 3201, 3427, 3768, 3821, 4157, 4366, 4411, 4621, 4996, 5107, 5343, 5552, 5765, 5963, 6105, 6350, 6525, 6720, 6841, 7105, 7366, 7422, 7618, 7808, 8032, 8209, 8541, 8746, 8827, 9123, 9283, 9423, 9772, 9953, 10041, 10207, 10587, 10766, 10935] 22

-1 should equal 198

[184, 364, 475, 747, 946, 1151, 1215, 1529, 1657, 1847, 2128, 2306, 2515, 2774, 2853, 3123, 3291, 3478, 3607, 3927, 4190, 4264, 4454, 4659, 4920, 5042, 5345, 5482, 5694, 5909, 6134, 6318, 6545, 6680] 92
-1 should equal 184

I suggest to add more fixed and random test sequences containing zero(s) and negatives, e.g.:

[0, 0, 0, 0, 1]       -->  -1  # none
[-1, 0, 1, 2, 3]      -->   0  # AP
[-1, 0, 0, 1, 2]      -->  -1  # none
[0, 1, 0, 1, 0]       -->   0  # none
[-1, 2, -4, 8, -16]   -->   1  # GP
[-1, -2, -3, -4, -5]  -->   0  # AP

etc.

subtract the bigger number with the smaller one.

bigger/smaller(a, b, c) or bigger/smaller(b, c) ?

Trivial arithmetic/conditions is neither novel nor interesting kata idea.