Repository navigation
feat(antigravity): route Gemini 4 Hydrogen's discovered xhigh tier through the CLI User-Agent - #8636
Conversation
Codex Review Summary: ✅ CompletedThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
✅ Deterministic PR hygiene checks passed. |
|
No actionable comments were generated in the recent review. Walkthrough
Merge Risk:
|
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e728583992
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if (!available.has(wireId)) return undefined; | ||
| suffixEffortMap[effort] = wireId; | ||
| } | ||
| if (available.has(`${pickerId}-xhigh`)) suffixEffortMap.xhigh = `${pickerId}-xhigh`; |
There was a problem hiding this comment.
Include xhigh in the downstream effort-family projections
When a user has saved gemini-4-hydrogen-xhigh in selectedModels (for example, while discovery exposed an incomplete family), discovering the complete ladder now collapses that row into gemini-4-hydrogen, but antigravityEffortFamilyIds() in src/providers/antigravity-effort-families.ts still enumerates only low/medium/high. Consequently, projectAntigravitySelectedModels() does not admit the base selection, and filterCatalogVisibleModels() filters out the newly collapsed model. If the old suffix row is retained, the public-collapse helper also fails to hide it. Extend the shared family enumeration to include an exact, discovered xhigh mapping so saved selections and public catalog collapse remain consistent with the parser.
Useful? React with 👍 / 👎.
…te Hydrogen's CCA evidence as unverified Review follow-up for #8636: - antigravityEffortFamilyIds now includes a validated optional xhigh tier, so an xhigh-only selection projects to the collapsed base, the retained -xhigh row is hidden, and disabled-tier inheritance counts xhigh. - The CLI User-Agent comments keep the observed Argon/Carbon evidence and describe Hydrogen's match as anticipated, not observed on CCA.
Maintainer judgmentMerging at head
|
Summary
Gemini 4 Hydrogen has a fourth reasoning tier,
xhigh. If Cloud Code Assist (Antigravity) listsgemini-4-hydrogen-{low,medium,high,xhigh}, OpenCodex today publishesgemini-4-hydrogen-xhighas its own picker row. That row also misses the Gemini 4 CLI User-Agent rule (/^gemini-4-<codename>(-low|-medium|-high)?$/from #8483), so the request goes out with the IDE User-Agent. CCA answers 404 to Gemini 4 models on that User-Agent (observed for Argon and Carbon).After this change:
-xhighsuffix. Argon and Carbon keep the CLI UA, and near misses (-max,-xxhigh,-tiered) keep the IDE UA.antigravityEffortFamilyIdsincludes the validated xhigh tier, so an xhigh-only saved selection projects to the collapsed base, the retained-xhighrow is hidden, and disabled-tier inheritance counts xhigh.<base>-xhighcollapses into one picker row with efforts[low, medium, high, xhigh]. Routing sendsxhighto the-xhighwire id. A three-tier family (Argon, Carbon) still clampsxhighto-high, and a lone-xhighinvents no ladder.xhighentry and still reads older snapshots.gemini-4-hydrogenfamily defaults tohigh, in both the catalog row and wire resolution without an effort.Evidence: Hydrogen's low/medium/high/xhigh ladder and its high default were verified live on Vertex AI by the author of #8583 (2026-11-13). Hydrogen has not been seen on CCA yet (the local CCA wire snapshot from 2026-11-11 lists only the Argon family). Hydrogen therefore stays out of the static Antigravity picker, like Carbon, and appears only when an account's CCA discovery lists it. Places where only Argon is registered (static picker, Argon quota group, Argon prices) are left unchanged. Code comments, the structure doc and the providers guide state this.
Verification
tests/adapters/google/gemini-4-hydrogen-antigravity.test.tsagainst theorigin/devversions of the changed source files: 6 fail, 1 pass (the unchanged three-tier clamp case). The xhigh family-projection case (selection, public collapse, disabled inheritance) fails without theantigravity-effort-families.tschange. With the change: 8 pass, 0 fail.bun run typecheck: no type errors.bun testprocess per file, all 0 fail: gemini-4-hydrogen-antigravity (8), provider-antigravity-effort-families (19), gemini-4-argon-antigravity (19), antigravity-discovered-families (11), google-antigravity-wire (62), antigravity-account-roster (14), antigravity-routing-carry (28), google-antigravity-replay (71), gemini-37-flash-migration (45), google-wire-shape (35), gemini-web-search (16), gemini-progress-history (8), google-models-listing (14), google-vertex-thought-signature (38), provider-antigravity-wire-snapshot (6), provider-antigravity-family-catalog (3), routing-capability-catalog (10), usage-antigravity-55 (5), model-rename-migration (56), server-images (82).tests/ci-workflows/file-size-ratchet.test.ts(9 pass;guides/providers.mdis at 1,999 lines),tests/test-layout.test.ts,tests/test-layout-tooling.test.ts,tests/ci-workflows/structure-ssot.test.ts: pass.bun run structure:check,bun run privacy:scan: pass.bun run testsuite locally. The host was shared with a concurrent full run, so this change is left to the CItest changedjob. No live CCA request was made (no account here lists Hydrogen).Checklist
Related: #8583 (Vertex Hydrogen registration), #8483 (Carbon CLI User-Agent).
Summary by CodeRabbit