codex smart model routing

Our most recent models already dynamically adjust their "thinking level" based on the complexity of the task. Do you have something else in mind?

Our most recent models already dynamically adjust their "thinking level" based on the complexity of the task. Do you have something else in mind?

@etraut-openai thanks for the quick reply. I just want it for sake of cost-efficiency, it might be not compute-efficient due to how infra is built but switching to lower price model for easier tasks during single codex prompt turn is good for cutting the cost.

The mechanism I mentioned should already provide you with cost savings for simple tasks because the model will consume fewer tokens for simple tasks. You can read more about that here.

The mechanism I mentioned should already provide you with cost savings for simple tasks because the model will consume fewer tokens for simple tasks. You can read more about that here.

thanks I have read it, I understand that reasoning mode is already smart routed for token reduction. My reasoning is basically purely based on pricing difference between different modelIds, for example(feel free to close this ticket as I think its not necessary/urgent):

GPT-5.2 Codex

Input tokens: ~$1.75 per 1M

Cached input: ~$0.175 per 1M

Output tokens: ~$14.00 per 1M

GPT-5.1 Codex Mini

Input tokens: ~$0.25 per 1M

Cached input: ~$0.025 per 1M

Output tokens: ~$2.00 per 1M

This feature request hasn't received enough upvotes, so I'm closing it.