[johnciavarella]

Problem

The #92 review asked for the shared pricing change to go in its own pull request: #92 (comment)

For a model with a scheduled built-in price change, Settings → Model pricing has two problems:

• All-history saves are dropped. An "apply to all history" save at today's built-in price is treated as "use built-in" and dropped, so the earlier scheduled price comes back for usage before the cutover.
• Dated saves are compared with the wrong price. They are checked against the latest built-in price, not the one in effect from the save's start date.
  • Saving the earlier built-in price for a date before the change pins it permanently, even past the scheduled increase.
  • Saving today's price from a date before the change is silently ignored.

What changed

• Settings display: Settings → Model pricing shows the built-in price in effect now, so a model with a scheduled change shows the rate that applies today.
• All-history saves: a save equal to a scheduled model's current built-in price is kept as an override. It then applies before the scheduled change too.
• Dated saves: a dated save is compared with the built-in price in effect from its start date.
  • Saving that price keeps following the built-in, including any later scheduled change.
  • Saving today's price from a date before the change is kept as an override for that window.
• All history: an "All history" save always replaces the timeline with one flat price after the user confirms. Saving the earlier price shown before a change keeps it after the change as well, and a test pins this.
• Unaffected models: models without scheduled price changes, and every other Settings flow (restore, custom and retired models, discounts), behave as before.

Today this affects:

• Codex: gpt-5.6, gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna
• Cursor: gpt-5.6-sol

The Antigravity PR (#92) adds Gemini models with scheduled increases on 2026-12-01 and 2027-01-01, and they benefit from the same handling.

Related: #92

Validation

• python3 -m unittest discover -s tests: 2,247 tests, OK (17 skipped)
• Tests: 9 new tests in tests/integration/test_model_pricing_compat.py.
  • With token_meter/app.py reverted to main, 7 fail: all-history keep, dated follow-schedule, a dated save of today's price before the change, replacing an existing override, the Cursor path with restore and idempotent re-save, the displayed price in effect now, and re-saving the shown price before a change.
  • The unscheduled-model guard passes on both.
• py_compile: OK
• git diff --check: clean

Limitations

• Not reinstalled: I didn't run the installer for this branch. The live install here runs the feat(runtimes): add Antigravity runtime with Gemini pricing #92 branch.
• Possible merge conflict: this PR and feat(runtimes): add Antigravity runtime with Gemini pricing #92 both touch model_pricing_settings, on adjacent lines. Whichever merges second may need a one-line rebase.