Skip to content

Rename the Gemini 4 Argon key from gemini-4-argon-eap to gemini-4-argon - #65

Open
SkalskiP wants to merge 1 commit into
mainfrom
chore/rename-gemini-4-argon-key
Open

SkalskiP wants to merge 1 commit into
mainfrom
chore/rename-gemini-4-argon-key

Conversation

@SkalskiP

@SkalskiP SkalskiP commented Oct 13, 2026 •

Copy link
Copy Markdown
Collaborator

What

Mechanical rename of the model key gemini-4-argon-eap → gemini-4-argon, no data changes.

Argon is released, and gemini-4-argon is the id the Gemini API, the Roboflow google_gemini@v6 block and the Playground use. The -eap suffix from #64 recorded that the committed runs went through Google's early-access endpoint; that provenance note now lives in the models.yaml comment, so the key matches the public model id for every downstream consumer (the Playground's benchmarks/vlm-exam snapshot keys its model links on it: [internal link removed]).

  • 38 results/*gemini-4-argon-eap* files moved and their "model" field rewritten (5,046 lines).
  • models.yaml, text_release.json, tests/test_config.py updated.
  • web/benchmark_summary.json re-keyed (key / id only; scores, tokens, costs untouched).
  • Text leaderboard charts + text_manifest.json regenerated: the manifest fingerprints the text summary, which carries model keys, so summary --check fails until it is rebuilt. Rendered content is unchanged (charts show display names).
  • Inference hashes unchanged: inference_hash excludes the key.

Validation

  • ruff check / ruff format --check: pass
  • pytest: 336 passed
  • vlm-exam validate: PASS (pre-existing legacy warnings only)
  • vlm-exam summary --check: matches saved results and model configuration

Co-authored-by: Cursor cursoragent@cursor.com

Argon is released, and gemini-4-argon is the id the Gemini API, the
Roboflow google_gemini@v6 block and the Playground use for it. The -eap
suffix (#64) recorded that the committed runs went through Google's
early-access endpoint; that note moves to the models.yaml comment so the
key matches the public model id everywhere downstream.

Mechanical rename, no data changes: the 38 result files are moved and
their "model" field rewritten, the registry, text release inventory and
config test updated, web/benchmark_summary.json re-keyed, and the text
leaderboard charts and manifest regenerated (the manifest fingerprints
the text summary, which carries model keys). Inference hashes are
unchanged: they exclude the key.

Validation: ruff check and format pass; 336 tests pass; vlm-exam validate
passes (existing legacy warnings only); vlm-exam summary --check matches.

Co-authored-by: Cursor <cursoragent@cursor.com>
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.

1 participant