Skip to content

feat(pipeline): default OpenAI generation model to gpt-5.4 - #4

Open
anthonyadame wants to merge 1 commit into
mainfrom
feat/gpt-5.4-default
Open

feat(pipeline): default OpenAI generation model to gpt-5.4#4
anthonyadame wants to merge 1 commit into
mainfrom
feat/gpt-5.4-default

Conversation

@anthonyadame

Copy link
Copy Markdown
Contributor

Default OpenAI generation model -> gpt-5.4

Flip DEFAULT_OPENAI_MODEL from gpt-4.1 to gpt-5.4 for the decompose and
recompose generation pipelines (the OpenAI default when no --model is given).

This ports the validated GPT-5.4 default from the Kaizen-3C staging repair-loop work
into the public CLI. gpt-5.4 was validated there as the strong, low-cost default
(~3.3-5.7x cheaper than Sonnet with >= delta on the commit0 portfolio). Fully
overridable via --model.

Scope / safety

  • Only cli/pipeline/recompose_v2.py + cli/pipeline/decompose_v2.py constants change.
  • The gpt-5* reasoning-model API params (no temperature, max_completion_tokens)
    are already handled by the existing is_reasoning branch — no call-shape change.
  • Unchanged on purpose: review/specialist models (review.py, specialist_review.py)
    and the benchmark baseline (oneshot_baseline.py, kept stable for comparison continuity).
  • config.py does not shadow the constant (_ARGPARSE_DEFAULTS["model"] is None).

Caveat

gpt-5.4 was validated for the staging repair loop, not separately benchmarked for
this one-shot decompose/recompose pipeline. It is a reasonable default modernization and
fully overridable; a paid end-to-end smoke (kaizen recompose --provider openai) is
recommended before the next release.

Validation

  • pytest cli/tests/: 123 passed, 7 skipped (no regression).
  • py_compile clean; is_reasoning("gpt-5.4") -> True (correct routing).

CHANGELOG updated ([Unreleased] / Changed); version bump left to the release process.

🤖 Generated with Claude Code

Flip DEFAULT_OPENAI_MODEL from gpt-4.1 to gpt-5.4 for the decompose/recompose
generation pipelines (the OpenAI default when no --model is given). gpt-5.4 was
validated as the strong, low-cost default in the staging repair-loop benchmarks
and is fully overridable via --model. The gpt-5* reasoning-model API params
(no temperature, max_completion_tokens) are already handled via the existing
is_reasoning branch -- no call-shape change needed.

Intentionally unchanged: review/specialist models (review.py / specialist_review.py)
and the benchmark baseline (oneshot_baseline.py, kept stable for comparison).
CHANGELOG updated. 123 passed / 7 skipped (pytest cli/tests/); no regression.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Updates the Kaizen CLI’s symmetric generation pipelines to use a newer default OpenAI model when --provider openai is selected and no explicit --model is provided.

Changes:

  • Switch DEFAULT_OPENAI_MODEL from gpt-4.1 to gpt-5.4 in decompose_v2 and recompose_v2.
  • Add an [Unreleased] changelog entry documenting the default-model change.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File Description
cli/pipeline/recompose_v2.py Updates the OpenAI default model constant used by the recompose pipeline when --model is omitted.
cli/pipeline/decompose_v2.py Updates the OpenAI default model constant used by the decompose pipeline when --model is omitted.
CHANGELOG.md Documents the new default OpenAI model under [Unreleased] (note: now duplicates an existing [Unreleased] section later in the file).

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Comment thread CHANGELOG.md
Comment on lines +6 to +10
## [Unreleased]

### Changed

- Default OpenAI model for `decompose` / `recompose` generation is now `gpt-5.4`
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants