Comment by bredren | Hacker News Reader

Comment by bredren | Hacker News Reader

FWIW, AI is not entirely locked down in the Apple ecosystem. Sure, they control it but they've already built the foundation of a major opportunity for developers.

There's an on device LLM that is packaged in iOS, iPadOS and macOS 26 (Tahoe) [1]. They even have a HIG on use of generative AI [2]

Something like half of all macs are running macOS 26 [3] already, so this could be the most widely distributed on-device LLM on the planet.

I think people are sleeping on this, partly because the model is seen as under powered. But I think we can presume it won't always be so.

I've just posted a Show HN of app for macOS 26 I created that uses Apple's local LLM to summarize conversations you've had with Claude Code and Codex. [3]

I've been somewhat surprised at the quality and reliability of Apple's built-in LLM and have only been limited by the logic I've built around it.

I think Apple's packaging of an LLM in its core operating systems is actually a fast move with AI and even has potential to act as an existential threat to Windows.

[1] https://developer.apple.com/videos/play/wwdc2025/286/

[2] https://developer.apple.com/design/human-interface-guideline...

[3] https://news.ycombinator.com/item?id=46209081

I can second this. I am nearing launch on an app that uses both the new SpeechAnalyzer and on device LLM and it has met or exceeded my expectations. A longer context would always be nice but then I remember its running on a phone.

Thanks for the follow-on anecdote. I'd be happy to try out your app. Please email me when it is available: rob@contextify.sh.

Are you using to for speech-to-text/text-to-speech? I have been using SpeechRecognizer & SpeechSynthesizer and they have been pretty underwhelming.

How are you evaluating it against your expectations?

Lick your finger before you stick it in the air. Amplifies the signal.

I'd like to know this too. Whisper is hard to beat.

Don’t a lot of Android devices come with Gemini Nano on the device?

Probably not as many out there as there are Apple devices because it is only the high end ones at the moment. I don’t think they are that far behind in numbers though.

I'd be curious to see an estimate on the google side.

Here are some real rough estimates in Apple's ecosystem:

For macos alone the install base is something like 110-130 million, and only Apple Silicon macs can run the new model, so maybe 45 million active macs are updated to macos 26 and can run their model.

There are a bunch of details but of the iPhones out there that are new enough to run Apple Intelligence and have iOS 26, something like 220 million can.

For iPad same conditions but for iPados its something like 60 million.

So, something like 325 million active devices are out there ready to run LLM completion requests.

Also in Chrome.

I've tested almost every LLM which will work on a modern iPhone and Apples models are universally terrible in comparison to almost every open-weights model, they're so bad it's a joke amongst devs who work in this space.

The only thing it's useful is super basic tasks like sentiment classification, summarization (sort of), or stuff like, "Does this message contain toxic/bad language, answer yes or no only".

"existential threat to Windows" - from your lips to whoever's ear.

There is no major opportunity for developers on Apple's platforms when they can just rug pull you as they please.

Reply on news.ycombinator.com