Apparently gpt4.5 is absolute thrash. It's slightly better than gpt4 and gpt 4o, but 30 times more expensive to run the same input, and more expensive to run versus their reasoning models
https://twitter.com/user/status/1895207017587015960
It's also extremely slow while being extremely expensive, likely as a result of just the sheer dataset size. It's like they realized it was the wrong path forward halfway into training it and it may be why we started getting all the reasoning models as a new path forward, but just dumped it on people as a sunk cost to say "here's 4.5, stop bothering us about it and if you are stupid enough to use it just because it's 4.5, great"
The competition has clearly caught up and OpenAI is stuck with their pants down with a lot of money spent and not much to show for it, and all their best talent has left.