Artificial intelligence models that were promoted from Apple on performance

Apple has announced updates of artificial intelligence models that operate a set of Apple Intelligence features via iOS, MacOS and more. But according to the company’s private standards, models that are subject to the oldest models of competing technology companies, including Openai.

apple He said in a blog post On Monday, the Human Test classified the quality of the text resulting from the latest “Apple On-DeVice”-which operates unlawful on the products, including the similar iPhone-“” similar “, but not better than, from Google and alibaba with a similar size. Meanwhile, these laboratories themselves classified the most capable Apple, which is called “Apple Server” and is designed to run in the company’s data centers, behind the year GPT-4O.

In a separate test evaluation of the Apple models’ ability to analyze images, the Raters of Humans Fad the Llama 4 Scout model on Apple Server, according to Apple. This is a little surprising. On a number of tests, Lama 4 Scouts Worse performance than the leading models of artificial intelligence laboratories such as Google, Anthropic and Openai.

Standard results add credibility to the reports indicating that the APER AIP research section has struggled to catch the competitors in the Cutthroat AI race. AIP capabilities of Apple in recent years DisappointedThe promotion of Siri, the promised, was delayed indefinitely. Some customers A lawsuit was filed against AppleAccusing the artificial intelligence marketing company of its products that have not yet been delivered.

In addition to the generation of the text, the Apple On-Device, which is about 3 billion of the parameter, and engines such as summarizing and analyzing the text. (Parameters are almost compatible with problem solving skills in the form, and parameters with general parameters in general are better than those that have less parameters.) As of Monday, third -party developers can benefit from this through the framework of the basic models of Apple.

Apple says that both Apple on-Device and Apple Server are outperforming the use of improved tools and efficiency compared to its predecessors, and can understand about 15 languages. This is thanks to an expanded training data set that includes image data, pdf, documents, manuscripts, graphics, tables, and champions.