This Week's Sponsor:

Drafts, Tally, Terminology, Simple Scan

Quality Productivity & Utility Apps, Ready for OS 26, from Agile Tortoise

Notes on Early Mac Studio AI Benchmarks with Qwen3-235B-A22B and Qwen2.5-VL-72B

I received a top-of-the-line Mac Studio (M3 Ultra, 512 GB of RAM, 8 TB of storage) on loan from Apple last week, and I thought I’d use this opportunity to revive something I’ve been mulling over for some time: more short-form blogging on MacStories in the form of brief “notes” with a dedicated Notes category on the site. Expect more of these “low-pressure”, quick posts in the future.

I’ve been sent this Mac Studio as part of my ongoing experiments with assistive AI and automation, and one of the things I plan to do over the coming weeks and months is playing around with local LLMs that tap into the power of Apple Silicon and the incredible performance headroom afforded by the M3 Ultra and this computer’s specs. I have a lot to learn when it comes to local AI (my shortcuts and experiments so far have focused on cloud models and the Shortcuts app combined with the LLM CLI), but since I had to start somewhere, I downloaded LM Studio and Ollama, installed the llm-ollama plugin, and began experimenting with open-weights models (served from Hugging Face as well as the Ollama library) both in the GGUF format and Apple’s own MLX framework.

LM Studio.

LM Studio.

I posted some of these early tests on Bluesky. I ran the massive Qwen3-235B-A22B model (a Mixture-of-Experts model with 235 billion parameters, 22 billion of which activated at once) with both GGUF and MLX using the beta version of the LM Studio app, and these were the results:

  • GGUF: 16 tokens/second, ~133 GB of RAM used
  • MLX: 24 tok/sec, ~124 GB RAM

As you can see from these first benchmarks (both based on the 4-bit quant of Qwen3-235B-A22B), the Apple Silicon-optimized version of the model resulted in better performance both for token generation and memory usage. Regardless of the version, the Mac Studio absolutely didn’t care and I could barely hear the fans going.

I also wanted to play around with the new generation of vision models (VLMs) to test modern OCR capabilities of these models. One of the tasks that has become kind of a personal AI eval for me lately is taking a long screenshot of a shortcut from the Shortcuts app (using CleanShot’s scrolling captures) and feed it either as a full-res PNG or PDF to an LLM. As I shared before, due to image compression, the vast majority of cloud LLMs either fail to accept the image as input or compresses the image so much that graphical artifacts lead to severe hallucinations in the text analysis of the image. Only o4-mini-high – thanks to its more agentic capabilities and tool-calling – was able to produce a decent output; even then, that was only possible because o4-mini-high decided to slice the image in multiple parts and iterate through each one with discrete pytesseract calls. The task took almost seven minutes to run in ChatGPT.

This morning, I installed the 72-billion parameter version of Qwen2.5-VL, gave it a full-resolution screenshot of a 40-action shortcut, and let it run with Ollama and llm-ollama. After 3.5 minutes and around 100 GB RAM usage, I got a really good, Markdown-formatted analysis of my shortcut back from the model.

To make the experience nicer, I even built a small local-scanning utility that lets me pick an image from Shortcuts and runs it through Qwen2.5-VL (72B) using the ‘Run Shell Script’ action on macOS. It worked beautifully on my first try. Amusingly, the smaller version of Qwen2.5-VL (32B) thought my photo of ergonomic mice was a “collection of seashells”. Fair enough: there’s a reason bigger models are heavier and costlier to run.

Given my struggles with OCR and document analysis with cloud-hosted models, I’m very excited about the potential of local VLMs that bypass memory constraints thanks to the M3 Ultra and provide accurate results in just a few minutes without having to upload private images or PDFs anywhere. I’ve been writing a lot about this idea of “hybrid automation” that combines traditional Mac scripting tools, Shortcuts, and LLMs to unlock workflows that just weren’t possible before; I feel like the power of this Mac Studio is going to be an amazing accelerator for that.

Next up on my list: understanding how to run MLX models with mlx-lm, investigating long-context models with dual-chunk attention support (looking at you, Qwen 2.5), and experimenting with Gemma 3. Fun times ahead!

Is Apple’s AI Predicament Fixable?→

On Sunday, Bloomberg’s Mark Gurman published a comprehensive recap of Apple’s AI troubles. There wasn’t much new in Gurman’s story, except quotes from unnamed sources that added to the sense of conflict playing out inside the company. That said, it’s perfect if you haven’t been paying close attention since Apple Intelligence was first announced last June.

What’s troubling about Apple’s predicament isn’t that Apple’s super mom and other AI illustrations looks like they were generated in 2022, a lifetime ago in the world of AI. The trouble is what the company’s struggles mean for next-generation interactions with devices and productivity apps. The promise of natural language requests made to Siri that combine personal context with App Intents is exciting, but it’s mired in multiple layers of technical issues that need to be solved starting, as Gurman reported, with Siri.

The mess is so profound that it raises the question of whether Apple has the institutional capabilities to fix it. As M.G. Siegler wrote yesterday on Spyglass:

Apple, as an organization, simply doesn’t seem built correctly to operate in the age of AI. This technology, even more so than the web, moves insanely fast and is all about iteration. Apple likes to move slowly, measuring a million times and cutting once. Shipping polished jewels. That’s just not going to cut it with AI.

Having studied the fierce competition among AI companies for months, I agree with Siegler. This isn’t like hardware where Apple has successfully entered a category late and dominated it. Hardware plays to Apple’s design and supply chain strengths. In contrast, the rapid iteration of AI models and apps is the antithesis of Apple’s annual OS cycle. It’s a fundamentally different approach driven by intense competition and fueled by billions of dollars of cash.

I tend to agree with Siegler that given where things stand, Apple should replace a lot of Siri’s capabilities with a third-party chatbot and in the longer-term make an acquisition to shake up how it approaches AI. However, I also think the chances of either of those things happening are unlikely given Apple’s historical focus on internally developed solutions.

Permalink

Hands-On with Sound Therapy on Apple Music

I’ve always been envious of people who can listen to music while they work. For whatever reason, music-listening activates a part of my brain that pulls me away from the task at hand. My mind really wants to focus on the lyrics, the style, the mix – all distractions from whatever it is I’m currently trying to do. It just doesn’t work for me.

But under the right circumstances and with the right kind of music, you can create an environment that is conducive to focus. At least, that’s the idea behind Apple’s recent collaboration with Universal Music Group. It’s called Sound Therapy, a research-based collection of songs meant to promote not only focus, but also relaxation and even healthy sleep.

The effort comes out of UMG’s Sollos venture, a group of scientists and music professionals focused on the relationship between music and wellness. Founded in 2023, the London-based incubator has used its findings to put together a library of music that, as Apple says, “harnesses the power of sound waves, psychoacoustics, and cognitive science to help listeners relax or focus the mind.”

Read more

Google Brings Its NotebookLM Research Tool to iPhone and iPad

Google’s AI research tool NotebookLM dropped on the App Store for iOS and iPadOS a day earlier than expected. If you haven’t used NotebookLM before, it’s Google’s AI research tool. You feed it source materials like PDFs, text files, MP3s, and more. Once your sources are uploaded, you can use Google’s AI to query the sources, asking questions and creating materials that draw on your sources.

Of all the AI tools I’ve tried, NotebookLM’s web app is one of the best I’ve used, which is why I was excited to try it on the iPhone and iPad. I’ve only played with it for a short time, but so far, I like it a lot.

Just like the web app, you can create, edit and delete notebooks, add new sources using the native file picker, view existing sources, chat with your sources, create summaries, timelines, and use the Studio tab to generate a faux podcast of the materials you’ve added to the app. Notebooks can also be filtered and sorted by Recent, Shared, Title, and Downloaded. Unlike the web app, you won’t see predefined prompts for things like a study guide, a briefing document, or FAQs, but you can still generate those materials by asking for them from the Chat tab.

NotebookLM’s native iOS and iPadOS app is primarily focused on audio. The app lets you generate audio overviews from the Chats tab and ‘deep dive’ podcast-style conversations that draw from your sources. Also, the audio generated can be downloaded locally, allowing you to listen later whether or not you have an Internet connection. Playback controls are basic and include buttons to play and pause, skip forward and back by 10 seconds at a time, control playback speed, and share the audio with others.

Generating an audio overview of sources.

Generating an audio overview of sources.

What you won’t find is any integration with features tied to App Intents. That means notebooks don’t show up in Spotlight Search, and there are no widgets, Control Center controls, or Shortcuts actions. Still, for a 1.0, NotebookLM is an excellent addition to Google’s AI tools for the iPhone and iPad.

NotebookLM is available to download from the App Store for free. Some NotebookLM features are free, while others require a subscription that can be purchased as an In-App Purchase in the App Store or from Google directly. You can learn more about the differences between the free and paid versions of NotebookLM on Google’s blog.

Inside Airbnb’s App Redesign: An AppStories Interview with Marketing and Design Leads→

Last week, I was in LA for Airbnb’s 2025 Summer Release. As part of the day’s events, Federico and I interviewed Jud Coplan, Airbnb’s Vice President of Product Marketing, and Teo Connor, Airbnb’s Vice President of Design, for AppStories to talk about the new features and app the company launched. It was a great conversation that you can watch on YouTube:

or listen to the episode here:

Last week’s launch was a big one for Airbnb. The company debuted Services and reimagined and expanded Experiences. Services are the sort of things hotels and resorts offer that you used to give up when booking an Airbnb stay. Now, however, you can book a chef, personal trainer, hair stylist, manicurist, photographer, and more. Better yet, you don’t have to book a stay with an Airbnb host to take advantage of services. You can schedule services in your hometown or wherever you happen to be.

Experiences aren’t entirely new to Airbnb, but have been expanded and integrated into the Airbnb app in a way that’s similar to Services. Services allow you to get the most out of a trip from locals who know their cities best, whether that’s a cultural tour, dining experience, outdoor adventure, or something else.

Chef Grace explaining how to serve sadza.

Chef Grace explaining how to serve sadza.

While I was in LA, I prepared a meal alongside several other media folks from around the world. Our instructor was Chef Kuda Grace from Zimbabwe at Flavors from Afar. We made sadza with peanut butter and mustard greens and then sat down together to compare notes from the day’s events, tell stories about our dining experiences, and get to know each other better.

The evening was a lot of fun, but what struck me most about it was something we touched upon in this week’s episode of AppStories. The goal of Airbnb’s redesigned app is to get you to leave it and go out into the world to try new things. It reduces the friction and anxiety of taking the plunge into something new and emphasizes social interactions in the real world instead of on a screen. In 2025, that’s unusual for an app from a big company, and it was fascinating to talk to Teo and Jud about how they and their teams set out to accomplish that goal.

I like Airbnb’s redesigned app a lot. It’s playful, welcoming and easy to use. What remains to be seen is whether Airbnb can pull off what it’s set out to accomplish. It isn’t the first company to try to pair customers with local services and experiences. Nor is it Airbnb’s first attempt at experiences. However, the app is a solid foundation, and if my experience at dinner in LA was any indication, I suspect Airbnb may be onto something with Services and Experiences.

Disclosure: The trip to LA to conduct my half of this interview was paid for by Airbnb.

Permalink

Inoreader: Boost Productivity and Gain Insights with AI-Powered Intelligence Tools [Sponsor]

Inoreader, a powerful platform for content discovery and consumption, is becoming smarter with the launch of Inoreader Intelligence – a suite of AI-powered features designed to help users save time, gain meaningful insights, analyze, and act on valuable information faster. Whether you’re diving into daily news, tracking niche industry updates, or conducting in-depth research, Inoreader Intelligence transforms how you process information with the introduction of article summaries and Intelligence reports.

The first addition to the suite, article summaries, brings instant clarity to your content. A new Summarize button appears inside every article, allowing you to generate a quick overview, explore a specific aspect with customizable prompts, or even ask your own questions – all within the Inoreader interface. You’re in full control: tailor how the feature looks and behaves, set your preferred language, and define a prompt persona to suit your workflow. Summaries are fully searchable, saving you time not just in the moment, but also later.

With the newly introduced Intelligence reports, Inoreader levels up bulk content processing. Select multiple articles at once and run custom or predefined prompts to extract insights, compare sentiments, and create reports. Once generated, each report becomes a new article, ready to annotate, export, or share. Want to collaborate? Reports can be made public or shared privately across Teams.

Article summaries are available in Inoreader’s Pro, Custom, and Team plans, while Intelligence reports are included in the new Team Intelligence plan and offered as an add-on for Pro and Custom users. All plans include monthly Intelligence tokens, and if you prefer more control, you can connect your own OpenAI API key for added flexibility.

With Inoreader Intelligence, you don’t just collect information – you save time while understanding it faster and better. Explore article summaries and Intelligence reports today at Inoreader.com, or download the app from the App Store or Google Play to experience smarter reading!1

Our thanks to Inoreader for sponsoring MacStories this week.

  1. Intelligence reports will be released on Inoreader’s mobile apps later this year. ↩︎

Podcast Rewind: Folding Phones, Window Management, Unwind Goes Hollywood, Murderbot, and Movie Tariffs

Enjoy the latest episodes from MacStories’ family of podcasts:

Comfort Zone

Matt’s phone is folding more than usual, Niléane has blown up her window management system again, and the whole gang learns about themselves in the challenge.

MacStories Unwind

This week, Federico quizzes John about Airbnb’s media event and how it compares to an Apple event, and John shares a superhero TV show and TV deal.

Magic Rays of Light

Sigmund and Devon highlight the premiere of sci-fi series Murderbot, discuss the potential impact of U.S. tariffs on international film productions, and catch up on tons of Apple Original trailers.

Read more

After Years in the Lab, CarPlay Ultra Emerges

Image: Apple.

Image: Apple.

Almost three years ago, Apple offered a sneak peek at an elaborate new version of CarPlay that spread beyond the rectangle of most car infotainment systems to occupy the instrument cluster space in front of front seat passengers. As I said at the time:

It will be interesting to see how widespread the adoption of the features Apple demoed will be. The company listed 14 automakers like Land Rover, Mercedes, Porsche, Nissan, Volvo, Honda, and Ford that they are working with, but it remains to be seen which models will adopt the new CarPlay and how quickly.

Originally set to launch in 2024, Apple announced today that what is now called CarPlay Ultra is available for new Aston Martin orders in the U.S. and Canada and soon as a software update to Aston Martins with the carmaker’s “next-generation infotainment system.” Apple says Aston Martin support in other countries will follow over the next 12 months.

Image: Apple.

Image: Apple.

The difference between standard CarPlay and the Ultra flavor is that the new version takes over a driver’s entire dashboard and extends beyond traditional CarPlay features to vehicle-specific data and controls. As Apple describes it:

CarPlay Ultra provides content for all the driver’s screens, including the instrument cluster, with dynamic and beautiful options for the speedometer, tachometer, fuel gauge, temperature gauge, and more, bringing a consistent look and feel to the entire driving experience. Drivers can choose to show information from their iPhone, like maps and media, along with information that comes from the car, such as advanced driver assistance systems and tire pressure, right in the instrument cluster.

Drivers can also use onscreen controls, physical buttons, or Siri to manage both standard vehicle functions like the car’s radio and climate, as well as advanced, vehicle-specific features and controls like audio system configurations or performance settings, right from CarPlay, giving them a more fluid and seamless experience. CarPlay Ultra also introduces widgets powered by iPhone that perfectly fit the car’s screen or gauge cluster to provide information at a glance.

Although CarPlay Ultra looks great, one car maker is a far cry from the 14 automakers listed on a slide at WWDC in 2022. That’s not surprising given pushback from automakers like GM, which announced in 2023 that it was ending CarPlay and Android Auto support for its EVs, and resistance from the likes of Tesla and Rivian to add CarPlay in the first place. However, Apple clearly oversold what would become CarPlay Ultra in 2022 in a way that in hindsight now feels a lot like Apple Intelligence’s enhanced Siri demo at last summer’s WWDC.

Still, I’m glad to see CarPlay Ultra emerge from the labs, even if it’s in a car that few people can afford. Auto tech inevitably trickles down to ordinary cars, and I’m sure CarPlay Ultra will, too, although I expect it will be quite a while until then.

Apple Maps Adds Dining and Hotel Picks, Plus F1 Features

Image: Apple.

Image: Apple.

Yesterday, Apple announced a couple of new features that are now available in the Maps app.

The first is a new label that’s been added to restaurants and hotels in the U.S., which have been awarded MICHELIN distinctions. According to Apple’s press release:

Users can now view and search for MICHELIN-starred, Green Star, and Bib Gourmand restaurants — along with MICHELIN Key hotels — starting in the U.S., with support for additional regions coming in the future.

The new MICHELIN integration also allows users to filter based on MICHELIN ratings and make restaurant reservations and book hotels through the MICHELIN Guide app. Apple says additional rankings and guides will be available soon from The Infatuation and Golf Digest, with more sources coming later. You’ll also be able to book tee times on golf courses using Supreme Golf.

Monaco.

Monaco.

For F1 fans, Apple has given Monaco the a special Detailed City Experience that’s designed for people attending Formula 1 TAG Heuer Grand Prix de Monaco 2025 as well as fans following along from home.

On its UK press site, Apple announced that:

The new Detailed City Experience features custom-designed 3D Monégasque landmarks of iconic locations, including Casino de Monte-Carlo, Fairmont Monte Carlo, Hôtel de Paris Monte-Carlo, Yacht Club de Monaco, and the F1 Paddock Club. Dark mode gives users an evening view of Monaco in a moonlit glow that activates at dusk.

The Detailed City Experience also features amazing details for road markings, land cover, trees, and public transit routes, as well as helpful navigation details like turn lanes, medians, bus and bike lanes, and pedestrian crossings. It also features a windshield view for drivers, which shows a road-level view as a user approaches complex interchanges, making it easier to see upcoming traffic conditions or the best lane for an approaching exit.

Apple has also added a lot of race-specific details, highlighting the course, landmark turns, as well as 3D representations of stands, cars, and more. Famous F1 courses have been added as a Maps Guide to promote Apple’s upcoming F1 movie starring Brad Pitt, too.

I appreciate the growing catalog of original curated content accessible through Maps. Travel planning on the web is a messy process, but with guides and well-regarded editorial content embedded directly in Maps, it’s often much easier to find a restaurant, hotel, or activity that sifting through an endless list of Google Search links.