I really wish GPT-5.5 had a bit more "taste" in design and planning.
For everything else related to code, it's the best model.
I hope GPT-5.6 closes the gap.
It would feel more complete then.
For now, I switch to Opus 4.8/GLM-5.2 to fix design issues or when I plan.
Many users criticize GPT-5.5 for lacking design taste and producing uncreative or low-quality outputs that force workarounds like model switching, while others appreciate its usefulness for generating multiple design iterations.
No Digg Deeper questions have been answered for this story yet.
But in a weird way, it makes sense that we ended up with different models that specialize in different things, maybe through org alignment, taste, process optimization, etc.
This is why I continue to believe that the combination of AI models is a great strategy.
I really wish GPT-5.5 had a bit more "taste" in design and planning.
For everything else related to code, it's the best model.
I hope GPT-5.6 closes the gap.
It would feel more complete then.
For now, I switch to Opus 4.8/GLM-5.2 to fix design issues or when I plan.
On a serious note, I found it very useful to have it come up with ten Different iterations of something, and then pick the best one and go from there. Also, have it build designs based off of. Real objects, but make it go deep with it, make it become very descriptive of a real object and then a design from that. That makes it get really creative and kind of pushes it outside of its comfortable vibe coding default mode.
@omarsar0 5.5 almost never ends up in a corner saying 'Sorry I was overconfident about that' like Claude but it has a harder time understanding the nuances of a large diverse codebase
@omarsar0 I am turning GPT5.5 into an expert website designer lol, it's a lot of fun.
@omarsar0 the split between best coder and best planner is stable across releases, openai wins execution, anthropic wins judgment
@omarsar0 We’re still forced to route steps across different labs just to get an elegant UI output.
@omarsar0 for design I gave up trusting it, i just specifically guide it to my taste
@omarsar0 Need more taste
@omarsar0 Grok build is creeping.
@omarsar0 interesting that code and design still feel split across models. feels like opus carved out its whole niche off that gap
@omarsar0 I wish it hade only the 1M context window
@omarsar0 Model routing gets easier when each model owns a failure mode: code diff, visual taste, long plan, cheap batch. Switching by vibe is where the mess starts.
@omarsar0 it hits an IP classifier and just outputs slop when you try to do anything creative with it.
sorry everything has been thought of and this marginally adjacent and we don't want frivolous lawsuits.
@omarsar0 Every model seems to have its own strength. Do you think one model will eventually handle both well?
@omarsar0 Frustrating to hear it’s still lagging on design logic—code dominance alone isn’t enough for full-stack dev workflows. Curious if the gap is in architectural planning or UI generation specifically.
@omarsar0 how do u use gpt5.5 for research tho , i mean claude is so much better , what i am trying to say is claude actually grills your idea and show u the reality but gpt 5.5 feels like a dream where everything feels amazing