GPT-4o feels noticeably dumber than GPT-4 for a lot of questions. I tried GPT-4o, but switched back to GPT-4 recently. There's probably a reason they added a button to quickly re-send your message to another model, hopefully they notice people not using GPT-4o as much and work on making it smarter. The speed improvement is really nice though, and I'm sure the improvements in other (non-text) modalities are real.
This is how I feel each time they release a new version that promises to be both faster and at least as smart as the old version. To be fair, Turbo did get a lot smarter over time.