GPT-6 Luna Light vs Medium vs High vs XHigh on one computer-use task

Quick easy test. I just told gpt-6 luna to delete my images from the gpt library.

• light: couldn't do it, and made something up based on a previous run.
• medium: could do the job, but stopped while loading.
• high: could do the job properly. Pretty much same as openai announced btw.
• xhigh: did the job, but thought too much before doing everything. Maybe it could even have done the QA afterward, but I told it to stop and move on.

Btw, I need the select all button back. Hope they don't do some sketchy expansion that makes it harder for people to delete things.

This was crazy. gpt-6 luna high realized the batch workflow was slow, tried to find a smarter bulk-delete path, caught a fake selection method before confirming, and then fell back to the safer real-checkbox route.

Lol no select all button for you my horse. Go runn.