I got iPadOS 26 on my m2 iPad Pro and have been playing with it. Not a huge fan of iOS 26 but it’s actually pretty good on the iPad. I opened shortcuts just to see what I could do and saw the apple intelligence option. Idk if anyone else has figured this out yet but I created one. Selected the use on device model option and flipped the follow up option. I also selected the ask every time option for the request. It takes a moment for the first response and the screen doesn’t do anything until it finishes processing. Further responses use a type to Siri like interface. I can have full conversations with an apple on device LLM. You can’t see your prior messages but it does seem to have awareness of them until you close the shortcut. It’s surprisingly robust too.
I asked it if it was able to have full conversations and it replied yes and explained itself a little. Then asked about the Iran/Israel conflict and it gave a two page detailed response. I even asked it about recipes and gave it my likes and dislikes and surprisingly came back with a well formatted recipe that looks legit. I asked 5 or 6 more questions and it continued to follow prior instructions and gave good answers. It seemed to try and format further answers but I think a bug may have broken it because other answers after the first had what looked like formatting characters that didn’t trigger it to display in a way that was clean. It was all one paragraph.
Very conversational, very friendly, and actually reminded me of GPT 4o without being sycophantic. That may change though as I use it more.
Thru further interactions it seems it can code, write full documents, and more. The list it provided is pretty extensive.
I asked it if it was able to have full conversations and it replied yes and explained itself a little. Then asked about the Iran/Israel conflict and it gave a two page detailed response. I even asked it about recipes and gave it my likes and dislikes and surprisingly came back with a well formatted recipe that looks legit. I asked 5 or 6 more questions and it continued to follow prior instructions and gave good answers. It seemed to try and format further answers but I think a bug may have broken it because other answers after the first had what looked like formatting characters that didn’t trigger it to display in a way that was clean. It was all one paragraph.
Very conversational, very friendly, and actually reminded me of GPT 4o without being sycophantic. That may change though as I use it more.
Thru further interactions it seems it can code, write full documents, and more. The list it provided is pretty extensive.
Last edited:
