A year after rolling out Apple Intelligence, Apple is heading into its annual developer conference with little to show for its AI ambitions, according to insider Mark Gurman. Instead of real progress, most of what’s coming is a mix of minor updates and fresh branding.
The headline AI announcement at this year's WWDC, set for June 9, is the opening up of Apple's own Foundation Models for third-party developers, Gurman says. These models run right on Apple devices and have about three billion parameters—small by industry standards and not especially powerful. Developers will be able to use them for simple features like text summarization in their apps.
Apple actually has much bigger and more capable AI models in-house. Gurman reports that Apple is already running models with 3, 7, 33, and 150 billion parameters. According to internal benchmarks, the largest cloud-based model performs at a level on par with current versions of ChatGPT. It can handle more complex tasks and, according to Apple employees, is a major step up from the smaller on-device models.
But the 150B model is only available for internal testing—mainly in Apple's "Playground" system, where employees can pit it against tools like Gemini and ChatGPT. There's no plan to launch it publicly.
Gurman says Apple leadership is still divided on strategy and worried about hallucinations, so for now, the company isn't building its own chatbot. That likely explains why Apple Intelligence now includes ChatGPT as a built-in option.
Besides opening up its models, Apple is expected to roll out a handful of smaller AI features: an AI-powered power-saving mode, a revamped translation app with Siri and AirPods support, and a rebranding of existing tools in Safari and Photos as "AI-powered." Gurman calls these changes more about marketing than real innovation.
Big AI projects are still on hold
Several of Apple's major AI efforts are running behind schedule. A new Siri with conversational abilities—meant to rival ChatGPT's Advanced Voice—is supposed to replace the current assistant, but its launch has been pushed back indefinitely. The Shortcuts app is also getting an Apple Intelligence overhaul for smarter automation, but it won't arrive before 2026.
Another project, the AI-driven health service "Mulberry," aims to let a revamped Health app answer medical questions and analyze health data. But it's hitting similar technical roadblocks as Siri. Apple is also experimenting with an internal web chatbot called "Knowledge," modeled after services like Perplexity, but it's nowhere near ready for release.
Apple was late to the generative AI party, debuting Apple Intelligence in June 2024. The company billed it as "AI for everyone," but the first beta fell flat with developers and users. Features like Genmoji, automatic text entry, and contextual notifications didn't stand out.
Meanwhile, Google is moving quickly with multimodal search and video AI, and OpenAI has teamed up with Apple's former design chief Jony Ive. Apple's approach to AI remains cautious. Even the "Personal Context" feature planned for Siri—which would let it use your personal data for more tailored answers—still isn't ready.
