@ios_dev_alb Not only tried it but also wrote the most detailed article on it: azamsharp.com/2025/06/18/the…
@ios_dev_alb I read about about and forgot again 😬 What‘s the key benefit of using it or where would you recommend to apply it?
@ios_dev_alb Tried it as well but Xcode 26 is so buggy it crashes the whole system. The compiler doesn’t reallynunderstand what is happening.
@ios_dev_alb I tried a little bit of chatting in a session on Mac sim as well as on device. Worked fine.
@ios_dev_alb @heestand_xyz Free on-device inference, privacy with no API keys to manage, offline access (zero remote latency), maintain small app size, structured output, tool calling for agentic use cases, will only get better over time, & simple integration — here’s 145 examples: github.com/IvanCampos/Fou…
@ios_dev_alb Played around with it a bit, and it worked quite well for me 🙂
Working with Tahoe on a secondary Mac. It is very buggy and Xcode crashes all the time. If you’re willing to get past that the Foundation models are amazing. Key benefit is speed for me - I’ve been using Gemini 2.5 flash for a fast model and it’s faster than that with similar accuracy. And it’s free :)
@ios_dev_alb It is incredibly easy to use, and so powerful. Generable is so dang cool! I have some big plans for it.
It works fine but nothing impressive. MLX already exists for multiple months now and gives access to more powerful models. The good thing about it is more the Swift API that allows for easy structured generation and tool call, plus that fact that it is shipped with Apple Intelligence. It will make it more accessible for many developers.
@ios_dev_alb Yea, check it out:
@ios_dev_alb I used it to create recipes and it worked almost fine it is fast , easy and free, probably will test with real data to see if it works oke or not
@ios_dev_alb Foundation models will work on iphone 15 and lower?
@ios_dev_alb For me it’s the fact that Premier but my only way to get my life moderated 🕊️🌞✨💎🍀💪🏻
@ios_dev_alb I've tried add layer to moderate dirty words in my chat app. But take too long time to response for this usage. @azamsharp give me some tips to prewarm model. Not tried these tips yet
@ios_dev_alb @grok tell about foundation models framework in brief
@ios_dev_alb Yeah, been exploring it a bit lately. It’s got potential if you approach it with the right mindset. Still testing how well it scales across use cases though. What’s your take?
@ios_dev_alb Not yet but super interested after I learn swift a bit more
@ios_dev_alb I already wrote a library for myself that does this with more then just Apple models and works on any platform including Linux. I see little benefit in this, too limited.
@ios_dev_alb Great! I was really impressed because the day before WWDC, I was building a personal app that would run AI locally. It was a complete mess then. But when I tried the Foundation models, I was able to build the core backend of the model serving with just three lines of code.
@ios_dev_alb Structured output is a game changer, even if you don’t like the model this will bring structured generation to a lot of other libraries as well + the API is nice