So recently, Anthropic has been doing… more than questionable things. As Ben Davis put it “as Anthropic shoots a 12 gauge into their foot over and over and over again”. Recently, I got banned from Claude, with the only notice for me is an email from Stripe saying my Anthropic subscription is canceled and my app saying “This organization has been disabled”. Switching over to OpenAI, my overall experience with the models has been much better, except where the Claude models are so much more conversational than the GPT models.
With my short time of using the GPT models and the ChatGPT interfaces, I’ve learned that you should never use GPT 5.3 Instant. It doesn’t follow explicit instructions in the customization prompt, ignores me when I try to correct it, and completely just acts more arrogant than Anthropic. GPT 5.4 however is actually a good model. It runs well and reasons nicely, follows my explicit instructions, and just overall doesn’t suck. The usage limits are so much better than using Claude.ai, because I don’t have to use the mid-tier model while watching the usage counter.
Now GPT 5.4 does suck at frontend/UI, which sucks, but having GPT 5.4 implement the backend and then have Opus 4.7 implement the UI with my Cursor sub actually works, and then Composer can go review the project and fix small bugs.
Now I really hope that Anthropic gets their act together and does something normal for once, because this is just sad and pathetic. Their models are great (except when you need to do something actually big), but they are not.
Google, fix your tool calls and maybe people will actually use the model. Just fine tune it or something you have like clusters upon clusters of GPUs.