Scheduled tasks (routines): show and allow choosing the model per routine
Status Open
Maintainer reply None cached
Activity 6 comments · opened Jul 1, 2026
Type: Feature request
Area: Scheduled tasks (routines) in the Claude desktop app / Claude Code (scheduled-tasks MCP server).
Problem
There's currently no way to see or set the model a scheduled task (routine) uses:
list_scheduled_tasks(MCP) returns schedule, cron,enabled,lastRunAt… but nomodelfield.update_scheduled_task(MCP) has nomodelparameter → the model can't be changed via the API.- In the app, the model actually used only shows up inside each run's session config JSON (
AppData/Roaming/Claude/claude-code-sessions/.../<session>.json,modelfield). Changing the model on a session (or a fork) does not reliably repilot future scheduled runs.
Result: users can't audit or control the per-routine model choice. With 20+ routines this matters for cost (Haiku vs Sonnet vs Opus) and quality.
Request
- Display each scheduled task's model — in the routines list / task detail, and in
list_scheduled_tasksoutput. - Set / update it — a
modelfield onupdate_scheduled_task, and a model selector in the routine settings that persists to future runs.
Why it matters
- Cost tuning: route simple digests to Haiku, reasoning-heavy routines to Sonnet/Opus.
- Easy migration when a new model ships (e.g. Sonnet 4.6 → Sonnet 5): today there's no reliable way to switch a routine's model.
Related: #64843 (also area:routines — scheduled task sessions not shown under the Scheduled sidebar section).
Showing cached comments. Read the full discussion on GitHub ↗
5 Comments
Found 3 possible duplicate issues:
This issue will be automatically closed as a duplicate in 3 days.
🤖 Generated with Claude Code
Please keep this open — it's not a duplicate that can be closed by pointing at a fix, because none of the flagged issues were ever fixed:
NOT_PLANNED(auto-closed as stale for inactivity).DUPLICATEand locked.modelparam tocreate/update_scheduled_task); it was itself flagged as a duplicate of yet another cluster (#48976 / #36496 / #56164) and closed without shipping.So the underlying problem has been reported repeatedly and never resolved — it just keeps getting auto-closed as stale/duplicate.
It still reproduces on v2.1.156 (Claude Desktop). Concrete repro: I set 10 scheduled routines to Sonnet 5 via the app's model selector; 0/10 took effect — each routine's session config JSON (
AppData/Roaming/Claude/claude-code-sessions/.../<session>.json,modelfield) still shows the previously configured model at the next run.This report also adds an angle the prior ones didn't cover: expose the model read-only in
list_scheduled_tasksso users can audit which model each routine runs — today that's only discoverable by parsing internal session JSON under AppData.Keeping this as the single consolidated, still-reproducing report rather than letting it auto-close into the same graveyard.
I’m not sure this is an enhancement so much as it’s a bug report. I used to be able to select the model on my routines as of a week ago. Checking now, though, this functionality seems to have disappeared in both Claude Desktop (MacOS, in my case) and on the Web UI.
I can confirm there seems to be no way to pick a model and I'm confident it used to work. What's the point of routines if you can't even choose the model?
@tim-rue @CypherPoet @RubenARNAUD
It's back. (Did Anthropic just accidentally dropped it for a couple of weeks?)
<img width="1005" height="353" alt="Image" src="https://github.com/user-attachments/assets/9593ec52-b875-4e95-a245-162e598a1fb8" />