The new ChatGPT widget for iOS.

The most recent update to the ChatGPT app for iOS has introduced a new feature that will make those who rely on the company’s AI assistant for work-related tasks quite happy: a customizable widget that lets you control which shortcuts can be placed for convenient access on the Home Screen.

The new widget is called ‘ChatGPT shortcuts’, and it comes in two sizes – small and medium. Alas, the company didn’t add support for large widgets on iPhone yet, nor did they roll out an XL widget for iPad users. I hope more sizes will be coming soon. Unlike the existing ‘ChatGPT’ widget, the new one allows you to personalize which shortcuts are visible on the Home Screen. Once added, long-press the widget, select ‘Edit Widget’, and you’ll be presented with options to choose which buttons you want to show and whether labels should be displayed underneath them or not.

What’s especially nice about these widgets is that they cover the full spectrum of “ChatGPT surfaces” now offered by OpenAI: Chat, Work, and Codex Remote. For each, you can swap default buttons for the following ones:

• Chat
  • New chat
  • Camera
  • Photos
  • Dictation
  • Voice
• Work
  • New work chat
  • Work camera
  • Work photos
  • Work dictation
  • Work voice
• Remote
  • New remote chat
  • Remote camera
  • Remote photos
  • Remote dictation
  • Remote voice

I’m a huge fan of Codex Remote for iOS and iPadOS: as things stand right now in the AI industry, OpenAI’s Remote “app inside an app” for iPhone and iPad is the only intuitive, full-featured solution to start new chats in a remote agent running on macOS directly from iOS and iPadOS. (I am as surprised as everyone else that Anthropic, who were first to the “remote control” space, haven’t shipped a comparable experience in Claude for iOS. Supposedly, that is coming “soon”.)

What do you want to open – Voice or Voice?

Even more impressively, Remote voice (a version of ChatGPT Voice that runs in Codex on the Mac and lets you have bidirectional conversations from the iPhone while the agent is running on the Mac and orchestrating threads there) can now also be accessed from Control Center and the Lock Screen via a dedicated control. Unfortunately, this control has the same icon as regular ChatGPT Voice (which runs in the cloud, not on your Mac). I know: it can all be very confusing, and I hope OpenAI brings some clarity to these icons soon. On a similar note, I wish OpenAI’s iOS team also offered the widgets’ new shortcuts as proper actions in the Shortcuts app. Right now, these launchers are only available from the Home Screen.

The new customizable widgets are the latest entries in a larger push by OpenAI to improve the performance of Codex Remote on iOS and other design changes for iPhone and iPad users. The ChatGPT app recently received a redesigned ‘latest photos’ picker in Chat and Work modes (long-press the + button in the text box, and you’ll see), a tweaked reasoning selector, and the cloud browser we covered earlier this week on MacStories.

Earlier today, Apple unveiled the new generation of Mac mini and Mac Studio, featuring the latest entries in the Apple silicon family of chips: the M6, available in the Mac mini, and the M5 Ultra, exclusive to the Mac Studio. You can read more details about the announcement and related specs in John’s overview.

As someone who’s been working with an M3 Ultra Mac Studio (on loan from Apple) with 512 GB of RAM for the past year (plus two separate M4 Mac minis) with a particular focus on local AI models and performance gains enabled by Apple’s MLX framework, I obviously am very interested in these new machines. To give you some context: my current workspace for my upcoming iOS and iPadOS 27 review, which lives in Notion, is entirely managed by a series of local agents running the latest DeepSeek-V4-Flash via MLX on the Mac Studio, which continuously scan the project for new notes, sources, and research material that is automatically categorized and linked in the chapters that I’m writing. So, yes, I’m sure I’ll have more thoughts on the new Mac minis and Mac Studios soon. In the meantime, I thought it’d be fun to break down the local AI-related details from Apple.

First, the M6. While it may not be the powerhouse that the M5 Ultra is, it’s an interesting new chip because of its new 2 nm process and dual Neural Engine system:

M6 is built using cutting-edge 2 nm process technology, packing greater transistor density into a smaller die for a major leap in performance and power efficiency. M6 also introduces a Dual 16-core Neural Engine, providing up to 2x the peak compute over previous generations to make on-device AI workflows run even faster. System frameworks can automatically utilize both engines simultaneously, enabling applications to see faster model execution.

But that’s not all: the new 12-core GPU (which has two more cores than the M4’s old GPU) now features Neural Accelerators in each core for the first time on the Mac mini. The result, according to Apple, is “4x faster AI performance and 2x faster graphics than Mac mini with M4”. If these numbers hold up and scale linearly, a local Mixture-of-Experts model such as Qwen 3.5-35B-A3B, which would run at ~17 tokens/sec on average on a base M4 Mac mini with 16 GB of RAM, could realistically generate output at over 60 tokens/second with the base model M6 Mac mini.

Based on what we’ve seen so far, the one downside of the M6 Mac mini is that it does not have Thunderbolt 5 ports; those are exclusive to the M5 Pro model, also announced today. As Apple also notes, Thunderbolt 5 allows multiple M5 Pro Mac minis to run local AI models as a single cluster (for example, using something like exo). Alas, that won’t be possible with the M6 Mac mini and its Thunderbolt 4 bandwidth bottleneck. Nonetheless, I’d be intrigued to compare the performance of the widely regarded Qwen3.8-27B and DeepSeek-V4-Flash models on the new M6 Mac mini.

Moving on to the flagship M5 Ultra Mac Studio, let’s start here:

With M5 Ultra, Mac Studio achieves up to 4.3x the peak AI compute performance of M3 Ultra and a staggering 9.8x more than M1 Ultra. Combined with up to 512GB of unified memory and 1.2TB/s of memory bandwidth, 50 percent higher than before, Mac Studio lets users run massive models entirely on device with complete privacy — without counting tokens or worrying about rising cloud costs.

As seen on the Mac Studio’s product page, the 512 GB RAM version of the M5 Ultra Mac Studio will only be available in “late October”. Unlike the M6, the M5 Ultra supports Thunderbolt 5, which will allow customers to run multiple Mac Studios in a cluster with shared memory to distribute inference across machines. And, according to Apple, a cluster of four Mac Studios yields three times the performance of a single Mac Studio.

Speaking from personal experience, I know that my M3 Ultra Mac Studio using oMLX can run DeepSeek-V4-Flash locally with generation averaging 35 tokens/second. Assuming a linear 4x increase, that would put the same model at over 120 tokens/second on an M5 Ultra Mac Studio. To put things in perspective, that kind of performance would be faster than any AI chatbot website, it’d be faster than many providers who offer a “fast” mode for their models, and it’d only be second to either dedicated NVIDIA PC clusters at home or specialized inference providers such as Cerebras or Groq…which are running in full-blown data centers. Sure, you would need a computer that is likely going to cost more than $20,000 to make it happen, but it’d still be possible on a single machine that is small, quiet, and that – in theory – any consumer can buy off the shelf.

It’s also interesting to consider the implications of M5 Ultra for running much larger models that would have typically required aggressive quantization to even stay in memory and produce any useful output. For example, Z.ai’s GLM-5.2 (a 744B MoE model) could run at up to 17 tokens/second on an M3 Ultra Mac Studio with 512 GB of RAM; the same person was even able to get the massive Kimi K3 model to run on a single machine…with decoding performance up to 3 tokens/second. You see where this is going: by continuing to push the envelope of what is possible with M-series chips, Apple is gradually making it possible to run frontier open-weight models on local hardware that takes up as much space as four Mac minis stacked together.

Obviously, for now, this is all theoretical: we’re going to need actual benchmarks to objectively measure the improvements and performance gains of the new chips. But if Apple’s claims hold up – and I have no reason to believe they won’t – I believe both the M6 and M5 Ultra will meaningfully shake up the landscape of local AI models on macOS, reinforcing, once again, that Apple continues making the best computers for AI in the industry.

I’ve recently been asked to explain what an “agent harness” is (particularly after I wrote about my favorite app of the year so far, Open Minis). I realized it was one of those concepts I could understand intuitively but not quite articulate. Thankfully, other people have done a better job of explaining it than I have.

Drew Breunig calls them “situated agents”:

Harrison Chase once excitedly shared an insight that agents are comprised of 4 things: a system prompt, a planning tool, a file system, and subagents. In the year-plus since he said that, I think this remains largely true. (Though you might tweak it to have general tools, etc.)

[…]

Imagine the simplest coding agent, and you at the keys. Let’s slowly zoom out and consider all the elements the harness can manage:

1. The Session: The current task and context, as both a trajectory and a durable, branchable log. You can zoom backwards, fork, and replay it.
2. The Environment: The instance, defined as a sandbox, terminal, worktree, computer, and/or container.
3. The Repo: The project. Versioned with Git, it contains your code, history, current work, AGENTS.md, guides, and hooks.
4. Memory: The person’s predilections, accrued over time, managing progress and past decisions.
5. Skills: The domain, artifacts describing reusable workflows or domain knowledge worth wielding in this situation.
6. The Team: Your colleagues and counterparts. Shared rooms, shared traces, project tracking tools, issues, and bug reports.
   • The Organization: The policies and audits, defined by legal, leadership, and procurement.
   • The Model: The LLMs themselves, the common artifact shared by all. Stochastic blobs we all poke trying to evoke positive outcomes. Log or train on their quirks and adjust.

I also thoroughly enjoyed this explanation and excellent visualization by Ted Spare, Dexter Storey, and Sarim Malik, writing for Rubric Labs:

A harness is the software that translates a model into a system that can affect its environment.

[…]

Functionally, a harness makes a model agentic, meaning it can take action.

And:

As coding agents begin to run for longer, and deploy more intelligence through dispatch, they are owning larger and more complex problems end to end, and the developer is less in the loop to steer the agent. The value of high quality planning increases as agents implement the plans more autonomously. Harnesses like Claude Code and Codex ship with a native planning mode, where the agent must first create a detailed Plan.md file with feedback from the user before executing. These harnesses then place a reference to the plan and the todo list into a top level state (system prompt) so that the agent doesn’t forget what it’s working on across long runs.

Given the multi-model, hybrid structure of the new Siri AI, we should probably assume Apple also made a lightweight “Siri harness” to aid the on-device orchestration of the entire system.

Ethan Millman, writing for The Hollywood Reporter:

Apple Music will soon launch labels on songs created with artificial intelligence, the company said in an email sent out to industry partners on Thursday.

In the email, obtained by The Hollywood Reporter, the streaming service said that AI music labels will come “later this year,” though Apple Music didn’t disclose a specific launch date.

The new labels come months after Apple Music launched transparency tags back in March, available for record labels and music distributors to disclose when content uploaded was “materially generated using AI.”

Good. As much as I enjoy working and tinkering with AI, AI “music” falls squarely in the category of AI implementations I do not understand and cannot even be remotely interested in. Perhaps I’m old-fashioned, but what’s the point of listening to “music” performed by something you can’t see live on a stage, or that you know has never lived a real life? (Obviously we can’t see The Beatles or Beethoven perform live today, but we know they were real humans with real motivations behind their work.)

If you ask me, AI music shouldn’t even be allowed on streaming services, but I suppose it’s too late to fix that problem by now. Hopefully Apple will also include a system-level toggle to permanently hide AI “music” from Apple Music’s UI, too.