One hundred passwords

><> (Fish), 95

Starting after the ' puts all characters onto the stack and at the end of the line it wraps around; closes the string with the same '; outputs the top of the stack as a number (n) and finish execution (;).

Characters for 13 and 10 were changed to space to not ruin the layout of the code (as fish is a 2D language.)

Some non-printable ASCII disappears in the answer. Correct program here.

'n;Y8Y9C-""NX(QN.`JY6%c--WP\1H  \LX `W&WX"9I;!H_PT%GM[140+M4@   &AET`NGPNa

Thanks for Martin Büttner and Sp3000 for improvements of 2 and 5 bytes.

Hex dump:

0000000: 276e 3b59 3859 3943 072d 2222 4e58 0e28  'n;Y8Y9C.-""NX.(
0000010: 5117 1a4e 2e08 600b 1c03 4a06 1759 3625  Q..N..`...J..Y6%
0000020: 0663 2d01 2d57 500c 5c14 3148 095c 0f4c  .c-.-WP.\.1H.\.L
0000030: 5803 0520 6057 2657 1f58 2208 3949 3b21  X.. `W&W.X".9I;!
0000040: 485f 5054 0b04 0b0b 2547 4d5b 3115 3430  H_PT....%GM[1.40
0000050: 2b0b 4d34 4003 0915 2641 4554 604e 0747  +.M4@...&AET`N.G
0000060: 504e 0361                                PN.a

CJam, 40

I just used a greedy approach (most of the time) to print the next possible number (carets mark correct program ends):

43;56) ; 34 K- 9+ 2* ;11;3Z+;54;6;1 ;80C+K- K+G-;32 6+ 7-Z+;73(  ; 11  60+K+;52;11;64;9;65;7 8;1;8  
 ^  ^ ^   ^^  ^  ^  ^  ^ ^ ^  ^ ^  ^  ^ ^  ^ ^ ^ ^ ^  ^ ^ ^  ^  ^   ^ ^^ ^ ^  ^  ^  ^ ^  ^   ^^^ ^  

The following numbers are printed correctly:

95 43 81 89 56 89 57 67 7 45 34 34 78 88 14 40 81 23 26 78 46 8 96 11 28 3 74 6 23 89 54 37 6 99 45 1 45 87 80 12 92 20 49 72 9 92 15 76 13 3 5 32 96 87 38 87 31 10 34 8 57 73 59 33 72 95 80 84 11 4 11 11 37 71 77 91 49 21 52 48 43 11 77 52 64 3 9 21 38 65 69 84 96 78 7 71 80 78 3 97

Test it here.

CJam, 56

Here is a somewhat more elaborate approach. It prints the last 56 numbers correctly:

0{;"箪⇓ⲩ䏨攛믹Ğᅏꛥ훻ᆾ㼖ꦅ땶읥湓ᠤ䡶"2G#b99bW):W=}:F;FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF

Chances are, SE has swallowed some characters, so here's a pastebin. You can try the code here.

It makes use of the fact, that the challenge asks for the programs in characters, so I'm using Unicode characters to encode the last 56 characters in a function, which then leaves the correct element on the stack by incrementing a variable on each invocation and selecting the corresponding element. The rest of the code just calls the function repeatedly.

The first correct program is the one ending at the second F in the code. I've fiddled around with the length of characters to be encoded for a while until I hit the right balance to make as much use as possible. One or two more might be possible with this approach.