Let individual (non-org) accounts set their own default model — for new chats AND scheduled tasks (Cowork)

For triage context: this was filed alongside two companion requests from the same Cowork workflow — #68925 (scheduled-task rename in place + adjustable toolbar order) and #68926 (expose context-window utilization to the model). Sharing the linkage in case they're useful to triage together.

TL;DR: As of desktop build 1.19367.0, per-task model selection persists and is honored by both scheduled runs and interactive chats — the only piece of this request still outstanding is an account-wide default model for individual (non-org) accounts.

---

Data point from an individual (Max) account, tested 2026-07-08. I ran a controlled test to see which parts still reproduce. Setup: a throwaway scheduled task whose only instruction was to report the model it's running as, with the per-task model dropdown set explicitly to Opus 4.8. Model confirmed via the session's model picker in each case (not the model's self-description).

| Check | Result |
|---|---|
| Per-task model selection persists after save/reopen | ✅ stayed on Opus 4.8 |
| Scheduled run honors the selection | ✅ ran as Opus 4.8 (picker + task output agree) |
| Interactive continuation of the task honors the selection | ✅ Opus 4.8, did not fall back to Fable |

Referring to the numbered Proposed Solution in the original post: items # 2 (make the per-task model an editable, persistent property) and # 3 (respect the task's model: for interactive chat instead of silently defaulting to Fable) look implemented in the current desktop build — the per-task dropdown now sticks and is honored end to end, for both the scheduled run and the interactive session.

The part of this request that still stands is # 1 — a per-account default model setting for individual (non-org) accounts. There's still no global default an individual account can set; org-level Settings → Models is Team/Enterprise-only. The per-task dropdown is a full workaround for one or two tasks, but for anyone running many scheduled tasks it means setting the model by hand on each one — that's the remaining gap worth keeping this open for.

Environment: Claude desktop 1.19367.0 (1a5be1), individual Max plan, tested 2026-07-08.

Individual (Max) account, macOS desktop 1.24012.9 / claude-code 2.1.219. Data for the piece @irishkid75th-create identified as still outstanding — the account-wide default — from a fleet rather than a single task.

I run 30 scheduled tasks, none of which has an explicit per-task model set. Reading what they actually ran (message.model + the sibling top-level effort key in ~/.claude/projects/**/<session>.jsonl, which records both for every past run):

• through 2026-07-24 — every task ran claude-opus-4-8 @ high
• from 2026-07-25 — every task ran claude-opus-5 @ high

All 30 changed model on the same day, in lockstep, with no edit from me. That is the default moving underneath a fleet, and it is the exact cost of having no individual-facing default setting: the one lever that governs 30 automations is the only lever I am not allowed to touch.

Two details that sharpen the request:

1. My chosen picker model has no influence. My interactive default is Fable 5; not one of the 30 tasks has ever run Fable. So "inherit what the user actually works in" is not the current behavior either — it is a separate, vendor-side default.
2. The rollover is unannounced and undiffable. Nothing in the app said the default had moved. I found it only by reading transcripts after the fact. For unattended automation, a silent change in both capability and cost profile across every task at once is the failure mode worth designing against — and an account-level default the user sets (and is notified about when it changes) fixes it at the root, where per-task pinning only fixes it one task at a time.