29
votes
Apple to partner with Google for Gemini access on iPhones, Apple Intelligence to power on device assistant
Link information
This data is scraped automatically and may be incorrect.
- Title
- Joint statement from Google and Apple
- Published
- Jan 12 2026
- Word count
- 98 words
Woah. I knew they were struggling with apple intelligence, but even so - unexpected.
I kinda saw it coming ever since they made the partnership with OpenAI (complex requests being sent to GPT), just didn't know with whom. At one point there were rumors they were going to buy Mistral.
But with hindsight now, it's natural that they partnered with Google. This is pure conjecture on my part, but here goes anyway:
Apple has been all about privacy for I-don't-know-how-many-years. Google on the other hand, is all about snooping around what you are doing.
Apple has access to telemetry as well, but it's not as rich, complete and large as Google's. There's a lot more devices using Android than iOS, and Google is far more invasive in their monitoring.
As a result, Apple simply does not have the same resources as Google to create their own LLM-driven assistant. Apple, being all about making the best of whatever feature/product they launch whenever they launch it, would obviously look for the best in the business that can make such a product. In this case, it's Google by a long shot.
It's not OpenAI because they do not have their own OS and devices, and it's not Microsoft either as their Copilot updates have proven to be unpopular in general (and copilot is basically GPT, it's not their own model built in-house from scratch). Claude and Mistral have the same problems as OpenAI, so they lack the data on what makes a good LLM assistant
It's smart to not join the lighting-yachts-full-of-cash-on-fire club when everyone else around you is doing so. You'll end up in a much better financial situation when the bubble pops.
Apple is a consumer electronics company that dabbles in software, and their biggest advantage is their custom SoCs, coprocessors, and large base of customers using their hardware. What it takes to train LLMs is massive amounts of commodity GPUs sitting in data centers, with a high CapEx and short usable life span.
Google owns data centers and sells cloud compute services. Apple's web services all run on Amazon AWS and Google Cloud Platform.
I personally prefer Apple's approach of making task-specific machine learning tools, rather than LLM mania. I don't need Siri to be a chat bot; I need it to be a reliable and fast voice control tool for CarPlay. Most of my issues with it are it not listening to me, lack of integration with Spotify and third party messaging tools, and the inherent awkwardness of talking to a computer instead of operating it with your hand. None of which will be solved by adding Gemini.
Yup same here. Back when they announced Apple Intelligence, I wrongly assumed that it was going to be a combination of different models running on device designed for specific tasks.
I use Siri in similar contexts, just to change music and set up reminders. For me a huge upgrade for it would be if it could remember context; for example, when I finish set up a reminder and notice that I did a mistake and want fix it, it would be able to understand that I’m talking about the last created reminder and edit it accordingly. As it is today, I would need to create a new one from scratch.
A small hope that I had at the time would be if these models would learn from our behaviors. For example when I set up reminders while I’m driving, I always set them to show up when I arrive at my destination. It would be cool if Siri could notice this pattern and would do that automatically whenever I create a reminder while driving.
But alas, it didn’t happen.
There's actually one bit of context that Siri remembers, and that's the name of someone you last sent a message to. If you tell Siri to "Tell Jimmy hi" and then "tell him to meet me at six today", it will know who "him" refers to.
That being said, it is the only example of this which I have seen.
My issue with Siri does relate to its conversational ability. Unlike the other assistants, it's the worse at parsing natural language. It often feels like I'm using a CLI app just with voice. A common example: during carplay, for some reason my podcast app (pocketcast) stops playing. I want to use Siri to continue playing it again.
If I say "Hey siri, play pocketcasts", it'll say, "Sorry, there is no track named "Pocketcasts" found. Would you like to search on the internet?".
Instead, I HAVE to say "Hey siri, RESUME pocketcasts".
Even if all they did was make Siri capable of asking clarifying questions when there's ambiguity (instead of hard failing) that would improve its usability dramatically.
I'll never forget the time I said "Siri, call FIRST_NAME" and it replied "Sorry, couldn't find a phone number for LAST_NAME". It had clearly found the contact info since it found the last name of that contact, but then it just broke. Could never repro it again.
Okay, that's just hilarious. Siri tried its best, and that's totally okay.
No, Apple's basically licensing the models to run them on their own servers.