[timacek]

Problem

Antigravity now advertises two Gemini families that the quota model registry does not know:
gemini-4-argon-{low,medium,high} and gemini-3.8-flash-{low,medium,high}.

Because getQuotaModelFamilyId returns the raw id for unknown models, each family falls through as
three unrelated quota entries:

• the account card lists three rows with raw ids (Gemini 4 Argon High, Gemini 4 Argon Low, ...)
  instead of one family row with the product display name, and each row gets its own percentage;
• the conservative family summary used for account comparison/recommendation counts each tier as a
  separate model, so one family can dominate the list and the aggregate is not comparable with
  Gemini 3.7 Flash, Gemini 3.5 Flash or Gemini 3.1 Pro;
• getAvailabilityModelCandidates has no candidates for the family id, so the live-availability
  badge cannot resolve physical ids for the aggregated row.

Both families draw from a shared upstream allowance (Argon is documented as one separate weekly-only
allowance; 3.8 Flash shares the Gemini pool), so the existing conservative family aggregation
(minimum remaining percentage, earliest reset) is the correct presentation for them.

Changes

• src/modules/cloud-account/utils/quota-model-families.ts
  • FAMILY_DISPLAY_NAMES: add gemini-4-argon → Gemini 4 Argon, gemini-3.8-flash → Gemini 3.8 Flash.
  • getQuotaModelFamilyId: register both families before the 3.7/3.6 Flash rule.
  • Both stay out of EXACT_PRESENTATION_FAMILIES: the families are rendered as one row, matching the
    documented shared allowance, rather than one row per physical tier.
• src/modules/cloud-account/components/CloudAccountCard.tsx
  • getAvailabilityModelCandidates: map the two family ids to their physical tier ids.
• src/tests/unit/quota-model-families.test.ts
  • aggregation, family-id and presentation coverage for both families.

Physical ids and display names were confirmed from a live Google AI Pro account (agy models).

Validation

Run on Windows 11, Node 24.22.0, pnpm 10.34.6.

• node ./node_modules/vitest/vitest.mjs run on src/tests/unit/quota-model-families.test.ts
  (vitest 4.1.8, harness pointed at this worktree because main currently fails pnpm run check:ci tests:
  the lockfile pairs vitest@4.1.8 with vite@5.4.21, and vitest imports vite/module-runner,
  which Vite 5 does not export. Both tests-unit jobs fail on main at that step).
  Result: 12/12 pass.
• Regression check: the same test file against the pre-change source fails 3/3 new cases and passes the
  9 existing ones, so the new assertions cover the behavior.
• prettier --check on the three changed files: pass.
• eslint on the three changed files: pass.
• tsc --noEmit: pass.

Limitations

• The exact quota-summary payload for Argon was not captured from a live account, so the change is
  verified at unit level plus the live model list. If Google reports the Argon allowance as a quota
  group instead of per-model entries, the existing group path already renders it and this mapping is
  unaffected.

• No packaged-app screenshot: the change is limited to family aggregation keys and has unit coverage.

• The change is focused and excludes unrelated edits.

• Relevant behavior and documentation have been checked for this change.

• The diff and attachments contain no credentials, private data or generated application artifacts.

[codereviewbot-ai]

Summary of Changes

LGTM. The pull request correctly registers and collapses the gemini-4-argon and gemini-3.8-flash quota families:

1. Family Mapping & Aggregation (quota-model-families.ts):

   • Added gemini-4-argon and gemini-3.8-flash entries with user-facing display names in FAMILY_DISPLAY_NAMES.
   • Registered prefix matching in getQuotaModelFamilyId prior to legacy version fallbacks.
   • Retained conservative family aggregation (omitted from EXACT_PRESENTATION_FAMILIES) to present shared allowance tiers as unified family rows rather than fragmented tier rows.

2. Availability Candidate Resolution (CloudAccountCard.tsx):

   • Updated getAvailabilityModelCandidates so aggregated family presentation keys map to their underlying physical tier candidates (-low, -medium, -high), ensuring live model availability and rate-limit indicators are properly surfaced.

3. Test Coverage (quota-model-families.test.ts):

   • Added unit test cases for tier aggregation, earliest reset calculation, display names, and presentation filtering for both model families.

---

🤖 Reviewed by codereviewbot.ai - Catch bugs before your team does.

[timacek]

Both tests-unit jobs fail at startup, before any test file is collected:

Error [ERR_PACKAGE_PATH_NOT_EXPORTED]: Package subpath './module-runner' is not defined by "exports"
in .../node_modules/vite/package.json imported from .../node_modules/vitest/dist/chunks/nativeModuleRunner.*.js

That is a dependency mismatch in the current lockfile (vitest@4.1.8 resolved against vite@5.4.21), and it
also fails on main without this diff (for example run 39636647774 at 86a3600). It is not caused by the change
in this PR.

How the affected behavior was verified instead, on Windows 11 with Node 24.22.0:

• src/tests/unit/quota-model-families.test.ts run with vitest@4.1.8 against this worktree: 12/12 pass.
• The same file against the pre-change source: 3/3 new cases fail and 9/9 existing cases pass, so the new
  assertions actually cover the mapping.
• prettier --check, eslint and tsc --noEmit on the worktree: clean. The formatting, lint and react-doctor
  jobs pass on this PR.