If this rumor is accurate, I’m certain there are compelling reasons behind this choice. Personally, I would prefer they select Anthropic. I suspect that one of the reasons, if not the only one, is that it is less capable of web searching compared to Gemini.
1 Like
This paragraph suggests to me that Apple may be taking a page out of Adobe’s book, i.e., don’t expend blood and treasure trying to build your own model if someone else offers a better one you can in turn offer to your own users:
Apple is said to be paying Google to create a custom Gemini-based model that can run on its Private Cloud Compute servers to power Siri. Gurman clarified that this doesn’t mean Siri will include Google services or Gemini features. Instead, Siri will simply be powered by a Gemini model in the background, enabling it to deliver the features that users expect with an Apple user interface.
One difference between Adobe and Apple on this front: Adobe didn’t ask Google to create a custom Gemini-based model to replace or enhance its own tool, Firefly. Adobe bills Firefly as being “commercially safe”—i.e., trained on licensed and public domain content and with built-in limitations on the kind of images that can be generated. It further will indemnify its enterprise customers against claims that their Firefly-generated content infringes on someone else’s IP. I suspect that they couldn’t provide a “commercially safe” product with a custom Google model.
1 Like
Hmmm … Looks like everyone’s doing it:
Microsoft will pay to use Anthropic’s technology for some AI features in Office 365 apps, according to two people involved in the effort. … While Microsoft’s use of Anthropic technology could be viewed as a negotiating tactic, leaders developing the Office AI features found Anthropic’s latest models simply performed better than OpenAI’s at automating tasks such as financial functions in Excel or generating Powerpoint presentations based on customer’s instructions, said one of the people who is involved in the effort.
OpenAI’s recent launch of its flagship GPT-5 model is a step up in quality, this person said, but Anthropic’s Claude Sonnet 4 performs better in subtle but important ways, such as creating Powerpoint presentations that are more aesthetically pleasing than what OpenAI’s models create, this person said.
From Microsoft to Buy AI From Anthropic in Partial Shift From OpenAI in The Information (via archive.today)
1 Like
If Microsoft is planning to “use Anthropic’s technology for some AI features in Office 365 apps” I imagine a lot of people would like to see Siri work with Anthropic. The same as people who use Google Workspace would like to see Siri work through Gemini to access their business apps.
We will just have to wait and see if Apple delivers a useful tool this time.
2 Likes
FYI
2 Likes
I had no idea archive.[fill in TLD here] was “infamous.”
1 Like
I have family who use Samsung phones and the first thing they do with a new device is disable Gemini because it is so bad. Apple shouldn’t have gone for the cheap option, Claude would have been a much better choice.
To be fair, it will be fine tuned by Apple running on Apple hardware.
I personally prefer Claude, but this likely won’t be plain Gemini.
Also FWIW when I want do research and comparison of options Gemini often does a better job than Claude. (For example looking for a service to receive emails for an expenses app I’m building. Gemini found Resend.)
Update: I was using Gemini to find a list quiz tools for my website and the results were mixed. What it found it reviewed in depth, but it missed some key contenders.
1 Like
I generally prefer Claude as well, but in my experience, all the bots have their own strengths and weaknesses.
For instance, most of the time I find ChatGPT’s over-eager insistence on offering me next steps that I didn’t ask for and don’t want very annoying, but now and again it does point out that I might want it to do something that I absolutely hadn’t thought of, but yes, it turns out that do want that thing. Claude and Gemini will suggest next steps if I ask, but don’t volunteer them if I don’t. For a certain (and maybe big) subset of users, unprompted next step suggestions might actually be quite a useful thing.
2 Likes
I sincerely hope so. My brother told me that when it’s enabled he cannot even crate a task or control his home, as it gets confused, and it works great if you turn it off.
I hope they improve it so it’s much better than the version trained for Android, but with Apple’s track record on AI I’m doubtful, especially if Google cannot pull it off themselves.
Apple has said we should expect the new Siri next spring. Maybe a startup the size of Anthropic couldn’t create a custom LLM designed to run on Apple’s PCC servers. And deliver it in time for Apple to integrate into their devices, while keeping their business running?
Interestingly, I asked ChatGPT, Claude, and Gemini (I use the paid versions of ChatGPT 5 and Claude 4.5) to conduct in-depth research and identify the AI that best suits my writing needs. Surprisingly, all three AIs selected Claude. However, other AIs ranked higher for different tasks.
While I’m using Claude for that reason, as well as other reasons I’ve mentioned earlier, it’s slower compared to ChatGPT and Gemini. Additionally, I occasionally encounter timeout errors with Claude that I don’t experience with ChatGPT or Gemini.
Apple’s track record on LLMs and voice-enabled digital assistants may not be sterling, but it has incorporated other flavors of AI into its devices with a greater degree of success—the computational photography built into the iPhone’s camera is a good example. Apple has done a good job of leveraging computer vision—a subset of AI—into a tool that makes it possible for just about anyone to take a legible photo without having to know much, if anything, about things like the exposure triangle.
Caveat emptor - It’s unclear from your comment, if you the tools to do a web search and then summarize the results for you. +100 Or If you asked them self aware types of questions.
For all of the language the vendors use around thinking, reasoning etc, these models are not self aware. Ars Technica has a good summary: Why it’s a mistake to ask chatbots about their mistakes - Ars Technica
…and if you want to dig deeper, there is an academic paper buried in the Ars Technica article.
(You may well already be aware of this, in which case, this was a Public Service Announcement.)
1 Like
I assure you, I have no illusions that they are “self-aware.” 
My prompt was essentially: Search reputable AI sources (tech journals, research institutions, professional reviews) to determine which model, ChatGPT-5, Claude Sonnet 4.5, or Gemini 2.5 Pro, is best for professional editorial work. Context: Editor for a writer producing books, blog articles, speeches, and formal reports. Consider voice and style preservation, editing without rewriting, editing long-form manuscripts, and maintaining theological/philosophical accuracy and nuance. Provide current expert consensus (2025), comparative strengths/weaknesses.
1 Like
FWIW, and I was curious. I copied your request into Raycast AI (don’t know which AI they selected). It’s response was as follows:
Best AI Model for Professional Editorial Work (2025)
Based on analysis of current expert reviews from reputable sources, Claude Sonnet 4.5 emerges as the strongest choice for professional editorial work with your specific requirements.
Comparative Analysis
| Feature | Claude Sonnet 4.5 | ChatGPT-5 | Gemini 2.5 Pro |
|---|---|---|---|
| Voice & style preservation | Best-in-class - “nails your writing style,” maintains author’s voice when given examples | Tends to “cut too much copy and lose important details” | “Too verbose and sterile” edits |
| Editing without rewriting | Superior - makes precise edits while preserving original intent | Over-edits, removes important content | Significant rewrites that alter tone |
| Long-form manuscript handling | Excellent - 1M token context window | ~400K token context window | ~1M token context window |
| Nuanced content handling | Strongest - “thoughtful analytical approach” for philosophical/theological content | Good for logical reasoning | Better for research/summarization than nuance |
Expert Consensus
-
Claude Sonnet 4.5 excels specifically for editorial work where voice preservation is critical. As noted in multiple reviews: “Claude captures writing style better than any other model—especially when fed examples of best work” and is “particularly effective for style tweaks and other changes to the tone of voice.”
-
ChatGPT-5 performs well for logical reasoning and technical content but tends to over-edit, “cutting too much copy and losing important details” in the process.
-
Gemini 2.5 Pro is strongest for research-heavy tasks with Google Workspace integration but produces edits that are “too verbose and sterile” for nuanced editorial work.
For theological/philosophical content requiring precision, Claude’s “safety-first design” and analytical approach provides the most reliable handling of nuanced concepts without distortion.
Recommendation: Use Claude Sonnet 4.5 with your best writing samples as reference material to maintain consistent voice and style across all editorial projects.
1 Like
Looks like there is a “consensus.”
2 Likes