I’ve been away from programming for two months, and I’ve returned to the current Usage split between Auto+Composer and API. I’ve been particularly active with Cursor over the past two days, using the Pro plan I purchased two weeks ago. In just one day, I’ve managed to complete as much work on the Pro plan as I used to complete on the Ultra plan in a single day. However, before this weekend, I’d already consumed half of the plan by interacting with GPT-5.4-high, unaware of the wonders of GPT-5.4-mini-xhigh.
If you don’t require continuous work on senior-level tasks, Composer 2 and GPT-5.4-mini can handle the rest. I’m using GPT-5.4-Mini-XHigh as a router that distributes tasks to GPT-5.4-Mini-XHigh, Composer 2, Gemini 3.1 Pro, and GPT-5.4 High through Subagents.
I haven’t brought the current product to the manual testing stage yet, but everything seems to be going pretty well.
