GPT-6 Sol is slow, bad at GUIs and forgetful for Mac app work, developer says
A Mac app developer says GPT-6 Sol is slow, gets stuck in local optimization loops, produces sloppy UIs and forgets instructions, with low cost as its only advantage.
I'm not sure what GPT-6 Sol is good at. But for my use case of building Mac apps, it doesn't seem to be very good. 1. It's slow. And because it's not very smart, it takes a lot of tokens & time to perform a simple task. Worse, sometimes it ends up in a local optimization loop and would stuck in this loop for a long long time. 2. It's terrible at building GUIs. I usually design a screen in Claude Design and let codex implement it. 5.6 Sol has been very good at this, especially with Computer Use it can produce near pixel-perfect UIs. GPT-6 Sol though produced the sloppiest UIs you can imagine - even with design reference and all kinds of tools/computer use for verification. 3. It forgets things. I instructed it to not do certain things, but because of the "local loop" in #1 above, it always forgets this instruction. And after a few loops, it starts to ignore my instructions entirely. On the positive side though, it's cheap. For simple and straightforward tasks, it still does a decent job. But then I'd use GPT-5.6 Luna for those tasks instead. What is your experience with GPT-6 Sol so far?
This post quotes another post: “GPT-6 Sol xhigh runs 5+ hours on 1-2% of usage limit, user says”