New opcodes in PR 5071 look interesting, but I think it is better to leave this experiment for a separate issue.

Complexities of PRs (ignoring tests, documentation, and generated files):

PR 4682:
 Include/opcode.h      |  12 +-
 Lib/opcode.py         |  21 +-
 Objects/frameobject.c | 241 ++++++++-----
 Objects/genobject.c   |  10 +-
 Python/ceval.c        | 426 +++++++++++------------
 Python/compile.c      | 884 +++++++++++++++++++++++++++++++++---------------
 Python/peephole.c     |  18 +-
 7 files changed, 979 insertions(+), 633 deletions(-)

PR 5006:
 Include/opcode.h      |   8 +-
 Lib/opcode.py         |  11 +-
 Objects/frameobject.c | 203 ++++++++-------------
 Objects/genobject.c   |  10 +-
 Python/ceval.c        | 344 ++++++++++++++++-------------------
 Python/compile.c      | 474 +++++++++++++++++++++++++++++-------------------
 Python/peephole.c     |  33 ++--
 7 files changed, 561 insertions(+), 522 deletions(-)

PR 5071:
 Include/frameobject.h     |   5 +-
 Include/opcode.h          |  18 +-
 Lib/opcode.py             |  25 +-
 Objects/frameobject.c     | 224 ++++++---------
 Objects/genobject.c       |  10 +-
 Python/ceval.c            | 507 ++++++++++++---------------------
 Python/compile.c          | 681 +++++++++++++++++++++++++++++++-------------
 Python/peephole.c         |  49 +---
 Python/wordcode_helpers.h |  33 ++-
 9 files changed, 827 insertions(+), 725 deletions(-)

PR 4682: +4 opcodes, -4 opcodes, and changed numerical values of 2 opcodes
PR 5006: +4 opcodes, -4 opcodes
PR 5071: +8 opcodes, -8 opcodes, and changed numerical value of 1 opcode

It looks to me that PR 5006 has the lowest complexity and is the most easy for reviewing.

I'm working on it since taking Antoine's PR 5 months ago.