Add a /slow mode that trades latency for discounted usage

What variant of Codex are you using?

Codex App and CLI (ChatGPT subscription usage)

What feature would you like to see?

I’d like a /slow mode that is symmetric with /fast, for long-running tasks where latency does not matter.

When /slow is enabled, Codex could use lower-priority or queued capacity—similar in spirit to the API’s Flex Processing tier—in exchange for consuming less ChatGPT Codex usage quota or fewer credits.

Example use cases

• An overnight refactor, migration, or test run
• Large repository analysis that can finish hours later
• Background tasks where cost/quota efficiency matters more than response speed

Expected behavior

• /slow toggles slow mode for the current thread or the next task, just as /fast enables faster processing.
• The UI clearly shows that the task may be queued or run with higher latency.
• The UI shows the reduced usage multiplier or credit cost before the user enables it.
• Model capability and reasoning quality stay the same; only scheduling priority and latency change.
• Slow tasks remain suitable for long unattended runs and do not time out merely because they are lower priority.

This would let users voluntarily give up speed when the task can run overnight, while stretching their included usage further.

Additional information

Related to #2916, which requests general OpenAI API service-tier configuration. This proposal is specifically for ChatGPT subscription users and asks for a simple /slow UX opposite /fast, without requiring an API key.

OpenAI’s Responses API already exposes a flex service tier for lower-priority processing, so a similar product-level option may be a useful precedent.

The issue names Codex App and CLI but no files or tests. Start by tracing the existing /fast behavior and how usage or service-tier information is exposed; done means a reviewed design and implementation for /slow that covers toggling, cost disclosure, scheduling, and unattended task behavior.