Usage dashboard 'Favorite model' shows wrong model on every time filter
Prior closure note: Previously filed as #55213, closed as a duplicate of #41471. That is incorrect. #41471 reports that session/token totals differ between the "All Time" and "30 days" filters — a data completeness issue. This issue is about the Favorite model stat specifically, which names the wrong model on every time filter. Different stat, different mechanism, different fix.
What's wrong
The "Favorite model" stat in the Usage dashboard Overview tab shows Opus 4.8 on the "All" filter.
Actual usage (from local session archive, cross-checked against journal DB, covering all available history):
| Model | All-time sessions |
|---|---|
| Haiku 4.5 | 452 |
| Sonnet 4.6 | 407 |
| Opus 4.5 | 112 |
| Opus 4.6 | 95 |
| Opus 4.7 | 95 |
| Sonnet 4.5 | 50 |
| Opus 4.8 | 29 — dead last |
Opus 4.8 is the least-used Opus variant and ranks last among all real models. It should not appear as a favorite on any time window.
Wrong on every filter
- All: Opus 4.8 shown — Sonnet 4.6 (407) and Haiku 4.5 (452) both vastly outnumber it
- 30d: Opus 4.8 shown — Sonnet 4.6 beats it 184 to 26
- 7d: Opus 4.8 shown — Sonnet 4.6 is the only model used in the past 7 days
Wrong even within the dashboard's own truncated dataset
The dashboard undercounts sessions significantly (646 shown vs 1,224 in the local archive — roughly half the actual history, consistent with the incompleteness reported in #55212). Notably, Opus 4.8 is still wrong even within that truncated view: Sonnet 4.6 and Haiku 4.5 both have many more sessions in any slice of the data.
Environment
macOS Desktop, current version, "All" filter selected.
Expected behavior
"Favorite model" should name the model used in the most sessions within the selected time window. On "All", that is Haiku 4.5 or Sonnet 4.6 by a wide margin.