>>512780687
The main difference is it's now a router that tries to figure out the best model/profile to route prompts to (Reasoning model vs chat model). You can still explicitly tell it which to use, but it's supposed to try to do that automatically. it's not too good at it and doesn't choose to use reasoning very often even in cases it probably should, which is why it can get simple math problems wrong like pic related.
The reasoning model is pretty good though and you can use it just by literally telling it to think harder.