You may recall how GPT-5.0 compared to Sonnet-4 seemed like a senior engineer vs a curious junior rushing on stuff It feels to me like GPT-5.2 compared to Sonnet-4.5 is ... kinda similar, except like 5x thinky spin What is amazing to me these days is the ability to handle a "monitoring loop" until done. We're... there now. We're in a world where LLMs can do it. Small example: codex-cli runs OH codereview-roasted and feeds back its output to the worker agent. (3+ CLIs use 5.2 and call each other via tmux sessions or an agent mail MCP) GPT-5.2 is able to figure out and run commands like

sleep 720

just like that when it needs it, and... things work

Nice, have you compared with Opus-4.5 at all?

Ah, no, I haven't! Practically not at all

I’ve recently switched

I'll love to hear your thoughts about its abilities / comparisons or anything

With the big caveat that I’ve not used it a whole lot, it seems much less likely to declare victory prematurely, which was Sonnet’s most annoying endearing trait.