Apple’s Latest AI Models Struggle to Match Competitors
Apple has updated the AI models behind its Apple Intelligence features, but internal benchmarks reveal they’re still lagging behind those from leading rivals like OpenAI, Google, and Meta.
In a blog post published on Monday, Apple revealed that human reviewers found the text quality produced by its new on-device AI model — known as Apple On-Device — to be on par with similar models from Google and Alibaba. However, its more powerful server-based model, Apple Server, ranked below OpenAI’s GPT-4o, which was released a year ago.
The results weren’t much better when it came to image analysis. According to Apple, testers preferred Meta’s Llama 4 Scout model over Apple Server, despite Llama 4 Scout often underperforming compared to models from Google, Anthropic, and OpenAI.
These benchmark outcomes support growing concerns that Apple is falling behind in the competitive AI race. While the company touts improvements in efficiency and tool usage, critics point to ongoing delays—such as the repeatedly postponed Siri upgrade—and even lawsuits accusing Apple of advertising AI features that aren’t yet available.
Apple On-Device, which runs directly on devices like the iPhone, includes around 3 billion parameters and powers features such as text summarization and analysis. Meanwhile, Apple Server operates in the cloud and is meant to handle more complex tasks. Both models now support around 15 languages and offer developers access through Apple’s Foundation Models framework.
Although Apple claims these AI tools now use a broader training dataset—including PDFs, charts, and infographics—the tech giant still trails behind more agile competitors in delivering cutting-edge generative AI features.
As Apple continues refining its AI efforts, consumers and developers alike will be watching closely to see if it can close the gap and turn ambition into real-world performance.https://www.bloomberg.com/news/features/2025-05-18/how-apple-intelligence-and-siri-ai-went-so-wrong
