[Bug]: Usage leaves Daybreak models unpriced because Codex and LiteLLM use different aliases

Before submitting

• I searched existing issues and did not find a duplicate.
• I included enough detail to reproduce or investigate the problem.

Area

apps/server

Steps to reproduce

1. Run a Codex session using gpt-daybreak-blue-latest or gpt-daybreak-red-latest.
2. Confirm the Codex rollout records that model name in turn_context.
3. Open T3 Code's Usage page and refresh the usage summary.
4. Find the Daybreak model in the model breakdown.

Expected behavior

Daybreak token usage should receive the corresponding API-equivalent price from LiteLLM:

• gpt-daybreak-blue-latest should use daybreak-blue-latest
• gpt-daybreak-red-latest should use daybreak-red-latest

The Usage page should include those records in its estimated cost and cache-savings totals.

Actual behavior

The token counts appear, but Daybreak records are classified as unpriced and contribute $0.00 to the estimated cost.

Codex records the operational model names with a gpt- prefix:

gpt-daybreak-blue-latest
gpt-daybreak-red-latest

LiteLLM publishes their rates under different keys:

daybreak-blue-latest
daybreak-red-latest

T3's Usage pricing performs an exact normalized lookup. It removes provider-path prefixes and normalizes casing, but does not reconcile these two upstream aliases.

This is separate from the model-picker classification fixed by #7659.

Impact

Major degradation or frequent failure

Cost totals are consistently understated for users with Daybreak activity. Token reporting still works, but potentially substantial usage appears as free.

Version or commit

main @ 45a2c4b2a

Environment

Codex app-server with Daybreak access; reproduced with gpt-daybreak-blue-latest on macOS.

Logs or stack traces

Codex transcript model: gpt-daybreak-blue-latest
LiteLLM rate key:       daybreak-blue-latest
Usage result:           unpriced / $0.00

Workaround

No in-app workaround. Users can see Daybreak token counts but must calculate their API-equivalent cost separately.

Start in apps/server and trace the Usage pricing lookup that normalizes model names, then compare its handling of the Codex Daybreak aliases with the LiteLLM rate keys described here. Done means both gpt-daybreak-blue-latest and gpt-daybreak-red-latest resolve to their corresponding unprefixed rates and appear in estimated cost and cache-savings totals.