›_ Codex Issues
Codex Issues / #36422

Docs ambiguity: scope of task-sensitive model/effort selection when settings are unpinned

Open 💬 0 comments Opened Aug 1, 2026 by jack1231222

What is the type of issue?

  • Documentation is confusing
  • Documentation may be incomplete

What is the issue?

The current Subagents documentation says:

If you don't pin a model or model_reasoning_effort, Codex can choose a setup that balances intelligence, speed, and price for the task.

This paragraph is shown for the App, CLI, and IDE surfaces, but it appears in the Subagents documentation and then refers to settings in an agent file. The intended scope is therefore unclear.

Could the documentation explicitly distinguish the behavior for:

  1. ordinary top-level Desktop/CLI/IDE turns;
  2. custom agents and spawned subagents;
  3. omitted model / effort fields in app-server turn/start;
  4. inheritance from a parent, thread, global default, or current UI selection; and
  5. any native task-sensitive selection performed when all relevant settings are unpinned?

An independent Windows Desktop sample with no model/effort pins observed the same Sol route for trivial and complex top-level tasks, and Sol with medium reasoning for trivial and complex spawned workers. This may be valid inheritance behavior, but users currently cannot tell whether it is expected, a staged rollout, or different from the documented task-sensitive selection.

This issue only asks for a precise documentation contract. The related product request for native per-thread Auto routing and atomic pre-submit application remains in #34278.

Where did you find it?

View original on GitHub ↗