The -O1 parameter seems useless and like it shouldn't be used because any conditional jump (except jcxz and loop*) that isn't explicitly marked "short" expands to a near conditional jump. With a "cpu 8086" directive the two-instruction workaround is used, with "cpu 386" the 386 single-instruction near jump is used.

Test case:

test$ cat test.asm

%ifdef CPU
        cpu CPU
%endif

backward:
        je backward
        jcxz backward

        jcxz forward
        jne forward
forward:

test$ nasm -v
NASM version 2.16.02rc2 compiled on Oct 12 2023
test$ nasm test.asm -l /dev/stderr -O1 -DCPU=8086
     1
     2                                  %ifdef CPU
     3                                          cpu CPU
     4                                  %endif
     5
     6                                  backward:
     7 00000000 7503E9FBFF                      je backward
     8 00000005 E3F9                            jcxz backward
     9
    10 00000007 E305                            jcxz forward
    11 00000009 7403E90000                      jne forward
    12                                  forward:
    13
test$ nasm test.asm -l /dev/stderr -O1 -DCPU=386
     1
     2                                  %ifdef CPU
     3                                          cpu CPU
     4                                  %endif
     5
     6                                  backward:
     7 00000000 0F84FCFF                        je backward
     8 00000004 E3FA                            jcxz backward
     9
    10 00000006 E304                            jcxz forward
    11 00000008 0F850000                        jne forward
    12                                  forward:
    13
test$

In the manual it is documented that it should work like this: https://www.nasm.us/xdoc/2.16.01/html/nasmdoc2.html#section-2.1.24

> -O1: Minimal optimization. As above, but immediate operands which will fit in a signed byte are optimized, unless the long form is specified. Conditional jumps default to the long form unless otherwise specified.

This wording was added in https://github.com/netwide-assembler/nasm/commit/dc0bf47feb96884d658d0eceba8ce669c81cf591

This is a bug in that we expect -O1 to optimise more than -O0, not to pessimise the code. If the behaviour is not changed I would advice at least adding a bigger disclaimer that -O1 is often worse than -O0.