Task subagent model: inherit ignores parent Grok, runs GPT-5.5 Medium

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Title: Task subagent model: inherit ignores parent Grok, runs GPT-5.5 Medium

Repro:

  1. Agent chat model = Cursor Grok 4.5 High
  2. Launch Task with custom agent (planner) that has frontmatter model: inherit
    (or omit model / pass inherit via orchestration criticalModel)
  3. Subagent card shows GPT-5.5 Medium

Expected: same model as parent (Grok 4.5 High / grok-4.5-xhigh)
Actual: GPT-5.5 Medium

Notes:

  • Cheap agents with explicit model: composer-2.5 work correctly
  • Parent picker clearly shows Cursor Grok 4.5 High
  • Docs: inherit = parent agent model

Steps to Reproduce

  1. On Windows, open Cursor Agent chat and set the parent model to “Cursor Grok 4.5 High”.
  2. Use a custom project subagent (e.g. .cursor/agents/planner.md) with frontmatter model: inherit.
  3. In that Agent chat, run a workflow that launches the Task tool for that planner subagent without pinning a different Task model (omit model / inherit).
  4. Observe the model label on the completed subagent card in the chat UI.

Expected Behavior

Per Cursor docs, model: inherit should use the same model as the parent agent. The planner Task subagent should run as Cursor Grok 4.5 High (same as the parent chat), not a different model.

Actual: parent chat = Cursor Grok 4.5 High; planner Task subagent with model: inherit ran as GPT-5.5 Medium. Explicit composer-2.5 Task calls in the same chat worked correctly.

Operating System

Windows 10/11

Version Information

Cursor: 3.10.17 (stable)
Commit: c89f45b831621cdc5e951694db44fecd8fab1150
Date: 2026-07-05
OS: Windows 10/11 (build 26200), x64

For AI issues: add Request ID with privacy disabled

Parent Agent chat: Cursor Grok 4.5 High
Task subagent (custom planner, model: inherit): GPT-5.5 Medium (unexpected)

Does this stop you from using Cursor

No - Cursor works, but with this issue

Replying to confirm issue, occured during plan
execution.
Not thrilled about getting hit on API pool
rather than first party.
Difference: My hit GPT 5.6 sol medium