Apple Is Pushing AI Into More of Its Products—but Still Lacks a State-of-the-Art Model

WIRED

4 hours ago

Jun 9, 2025 8:22 PM Apple took a measured approach to AI at WWDC. A new research paper suggests the company is skeptical about some recent AI advances, too. At WWDC25, Apple showed it was taking a more incremental approach to AI development. Courtesy of Apple
Apple continued its slow-and-steady approach to integrating artificial intelligence into devices like the iPhone, Mac, and Apple Watch on Monday, announcing a raft of new features and upgrades at WWDC. The company also premiered the Foundation Models framework, a way for developers to write code that taps into Apple's AI models.
Among the buzzier AI announcements at the event was Live Translation, a feature that translates phone and FaceTime calls from one language to another in real time. Apple also showed off Workout Buddy, an AI-powered voice helper designed to provide words of encouragement and useful updates during exercise. 'This is your second run this week,' Workout Buddy told a jogging woman in a demo video. 'You're crushing it.'
Apple also announced an upgrade to Visual Intelligence, a tool that uses AI to interpret the world through a device's camera. The new version can also look at screenshots to do things like identify a product or summarize a webpage. Apple showcased upgrades to Genmoji and Image Playground, two tools that generate stylized images with AI. And it showed off ways of using AI to automate tasks, generate text, summarize emails, edit photos, and find video clips.
The incremental announcements did little to dispel the notion that Apple is playing catch up on AI. The company does not yet have a model capable of competing with the best offerings of OpenAI, Meta, or Google, and still hands some challenging queries off to ChatGPT.
Some analysts suggest that Apple's more incremental approach to AI development is warranted.
'The jury is still out on whether users are gravitating towards a particular phone for AI driven features,' says Paolo Pescatore, an analyst at PP Foresight. 'Apple needs to strike the fine balance of bringing something fresh and not frustrating its loyal core base of users,' Pescatore adds. 'It comes down to the bottom line, and whether AI is driving any revenue uplift.'
Francisco Jeronimo, an analyst at IDC, says Apple making its AI models accessible to developers is important because of the company's vast reach with coders. '[It] brings Apple closer to the kind of AI tools that competitors such as OpenAI, Google and Meta have been offering for some time,' Jeronimo said in a statement.
Apple's AI models, while not the most capable, run on a personal device, meaning they work without a network connection and don't incur the fees that come with accessing models from OpenAI and others. The company also touts a way for developers to use cloud models that keeps private data secure through what it calls Private Cloud Compute.
But Apple may need to take bigger leaps with its use of AI in the future, given that its competitors are exploring how the technology might reinvent personal computing.
Both Google and OpenAI have shown off futuristic AI helpers that can talk in real time and see the world through a device's camera. Last month OpenAI announced it would acquire a company started by the legendary Apple designer, Jony Ive, in order to develop new kinds of AI-infused hardware.
Even if Apple still lags behind in terms of building advanced AI, the company is publishing AI research at a steady clip. A paper posted a few days before WWDC points to significant shortcomings with today's most advanced AI models—a convenient finding, perhaps, if you are still getting up to speed.
The paper finds that the latest models from OpenAI and others, which use a simulated form of reasoning to solve difficult problems, tend to fail when problems reach a certain level of complexity. The Apple researchers asked various models to solve increasingly complex versions of a mathematical puzzle known as the Tower of Hanoi, and found that they succeeded up until a point, then failed dramatically.
Subbarao Kambhampati, a professor at Arizona State University who previously published similar work on the limits of reasoning models, says Apple's research reinforces the idea that simulated reasoning approaches may need to be improved in order to tackle a wider range of problems. Reasoning models 'are very useful, but there are definitely important limits,' Kambhampati says.
But even if the work suggests that a more cautious approach to AI is warranted, Kambhampati does not believe Apple is being complacent. 'If you know what's going on inside Apple, they're still pretty gung-ho about LLMs,' he says.