When Codex encounters something requiring permission—such as internet access, writing outside the workspace, or running an elevated command—Approve for me sends that request to a separate reviewer agent. That reviewer makes an additional model call, consuming extra input and output tokens. OpenAI explicitly states that automatic review “uses extra model calls” and can add to Codex usage.
@SpikeRiser@rezoundous Hmmm. I was wondering why I was being charged for auto-review when I have it shut off in settings. Kinda curious of the same… why does “approve for me” cost more usage? seems kinda lame. I don’t think we should have to “pay” to use approve for me.
Because “Auto-review” in Settings is not actually a global kill switch for existing Codex chats. The naming is confusing.
OpenAI documents that the Settings → General → Auto-review toggle merely makes the “Approve for me” permission mode available. It does not select that mode or change the permission mode of an existing chat. Therefore, a Codex chat that was already using Approve for me can continue generating auto-review usage even after you shut the Settings toggle off.
@rezoundous The days of resets are apparently over. I will use my next reset more wisely. I got greedy and was banking on them continuing to give free resets. Lesson learned.
@Teknium@daniel_mac8@arena Opus 5 is a piece of shit. No nice way to put it. If you want bugs in your code and it to tell you it’s writing them, by all means use it.