There's a new AI model out every other week, so here's how I check whether it's worth upgrading or not.

There's a new AI model out every other week, so here's how I check whether it's worth upgrading or not. GPT-5.5 just came out, so let's see how much better it is than 5.4. I usually always do three tests: a UI build, a debugging task, and an agentic task. So for test one, I asked them to build a Reddit-style feed. And as you can see, they both shipped working UIs and look pretty decent. So the differences are very subtle. 5.4 built exactly what I asked for, and 5.5 just went a little bit further for that extra mile. The test two, I planted five bugs in this really old gRPC project. I added some bugs that break tests and some that don't. I just asked them to figure out what was going on and fix them. 5.4 found three out of five, which were the ones that break tests. 5.5 found four out of five, so the same three that break tests, plus a wrong version string that I picked up on from going through the whole repo. So I think the pattern is very similar here. 5.5 seems to keep looking after the obvious problems are fixed, which is nice. For test three, I asked them to add a logging interceptor to that same repo. So they both implemented what I asked for with passing tests, but 5.5 made a few more notes on its reasoning. Like I said, it's an agentic task, so it noted that standard logging library wasn't available for my Go version, and thus avoided it. But then 5.4 matched the repo's conventions for file placement and 5.5 didn't. So I would say test three is mixed, but in my opinion, 5.5 just kept me involved the whole time. So in conclusion, if you're mostly chatting, writing short scripts, you probably won't notice a big difference. But if you're doing multi-step work where you want the model to read the freaking room and catch constraints, and very importantly, stop and ask before doing, I think that's where 5.5 pulls ahead, and that's the version that I reached out for now. What do you think? Byeee~