Add an explicit Smart Mode for automatic model and reasoning-effort selection

What variant of Codex are you using?

App

What feature would you like to see?

Codex currently exposes several model and reasoning-effort options, such as Medium, High, Extra High, Max, and Ultra. However, most users cannot reliably determine which level is appropriate for a specific task.

Although Codex may automatically balance model choice, reasoning quality, speed, and cost when these settings are not pinned, this behavior is not sufficiently visible or understandable in the user interface.

Please add an explicit “Smart Mode (Recommended)” to the model and reasoning selector.

Expected behavior:

1. Automatically classify the task by complexity, risk, context size, required tools, and validation requirements.
2. Select the appropriate model and reasoning effort for each task or task phase.
3. Use faster and more economical settings for simple searches, summaries, status checks, and narrowly scoped edits.
4. Automatically increase reasoning effort for architecture design, complex debugging, financial or legal logic, database migrations, security reviews, deployment planning, and rollback analysis.
5. Display the selected model and reasoning level, with a short explanation such as:

“Sol / High selected because this task involves a multi-file database migration and financial controls.”

1. Allow the user to override or pin the selection at any time.
2. Never silently reduce reasoning quality during high-risk phases merely to optimize cost or latency.
3. Record the effective model and reasoning level in the task or audit history.
4. Re-evaluate the level when a task moves between phases—for example:

research → implementation → testing → migration → production deployment.

1. Ask for confirmation before switching into an unusually expensive or long-running mode such as Max or Ultra.

Suggested selector design:

• Smart Mode (Recommended)
• Fast
• Standard
• Deep
• Manual Settings…

Example:

A routine file search could use a faster model with Medium reasoning. A complex accounting-system migration could automatically use Sol / High. A production database migration, statutory tax calculation, or rollback review could escalate to Sol / Extra High.

This would make Codex easier for ordinary users while preserving control for advanced users. Users should describe the outcome they need; they should not have to understand model architecture or estimate reasoning complexity before every task.

Additional information

_No response_