One more byte for lgdt instruction ?

Until yesterday I was using Fasm 1.56 for my compilation issues it works fine (because programs were running as I expected them to). But something strange appears when I decided to move to Fasm 1.65.17. The compilation was done without error but the program hang.So I ask my self the question WHY ?

1) I though I made some change and forgot about it so I looked into my code and compare it with the one some week before (I use to backup project like that).There were no change.

2) I fired again Fasm 1.65.17 and load my program into a debugger and begin tracing, I came accross this instruction

Code:

 lgdt fword [gdt10]
 ...
 gdt10 dw ?
       dd 0x92000000
       dw ?

the code above disassemble like this:
66 0F 01 16 ....

the result was a wrong address in GDT base address

Note:
I did not want data override byte because the code was under a 'use16'

Result 1: Strange byte 0x66

3) I fire this time with fasm 1.56
There was no 0x66 byte and GDT was just as I expected it
I also notice that the result file size after 1.65.17 compilation was one byte more than 1.56 compilation (probably this 0x66).

4) Question 2: Was I wrong since the begining (bug corrected in 1.56 lead to 1.65.17) ?
I wrote this just to be sure

Code:

 use16
 align 1
 org 0
 lgdt fword [gdt10]
 lgdt fword [gdt10]
 db 'G'
 gdt10 dw 0x55
       dd 0x77777777

here are the results:
from fasm 1.56 17 bytes

Code:

 0F 01 16 0B 00
 0F 01 1E 0B 00
 47
 55 00
 77 77 77 77

from fasm 1.65.17 19 bytes

Code:

 66 0F 01 16 0B 00
 66 0F 01 1E 0B 00
 47
 55 00
 77 77 77 77

My question now is What is the correct version ?
Bug or not ?

Code: